Omar SobhandClaude Opus 5
a2d7e3ea92
The boot log has said `continuous_research — 4 role skills bound, 8 unresolved`
on every start. Those eight roles ran without the instructions their template
promised them, silently: `skills_loader` reports the miss and carries on.
Worse than the missing files was what the prompts described. v1 told the
harvester to sweep "RSS feeds, GitHub trending, HN front page, YouTube /
podcast RSS" — none of which exist. `harvest.rs` searches arXiv and nothing
else. A role prompt describing a machine that was never built is the failure
this codebase keeps paying for, so v2 describes the machine that exists.
Roles now match the pipeline: paper_reader (the harvest already ran; read the
manifest and the papers), signal_ranker (unchanged in spirit), script_writer
(the podcast half, which had no role at all).
Seven skills authored under skills/research/, kebab-case to match the loader —
team_template_loader.rs:177-181 documents the snake_case/kebab-case trap that
already unbinds skills elsewhere:
arxiv-daily what the harvest guarantees, so an agent does
NOT re-search arXiv and corrupt the seen-set
paper-to-project-relevance name a file or roadmap item, or say "no bearing"
duplicate-detection the seen-set catches identity; this catches the
same work under a different id
signal-to-noise-ranking novelty/relevance/depth, and the two biases to
resist (recency up, inconvenience down)
executive-summary-writing what it is / why it matters / what to do — decide,
do not hedge
obsidian-vault-conventions the vault is a human's live workspace; never
main, never reorganise, hash the body not the file
podcast-dialogue-writing write for someone on a treadmill; the 10-70 char
highlight bound is the API's, not a style rule
`web_fetch` dropped from mcp_bundles: runtime_provision.rs binds every mission
claw to `["clawmates_door"]` and never reads that field, so declaring it
instructed roles to use a tool that never arrived. The prompts say `curl` via
Bash, which is what they actually have.
Boot now reports `continuous_research — 11 role skills bound`, with no
unresolved clause. 344 tests pass.
Co-Authored-By: Claude Opus 5 <[email protected]>
2.1 KiB
name, description, when_to_use, tags
| name | description | when_to_use | tags | ||
|---|---|---|---|---|---|
| duplicate-detection | Recognising that two harvested items are the same work, and why the seen-set is not enough on its own. | You are processing a harvest manifest and must avoid covering the same work twice. |
|
The same work wearing different clothes
The harvest already dedupes by identity: corpus_items keys papers by
arxiv:NNNN.NNNNN with the version suffix stripped, so 2401.12345v3 and
2401.12345v1 are one paper and a paper already shelved never reappears. That
is exact-identity dedup, and it is done for you.
What it cannot catch is the same work arriving under a different identity.
The cases the seen-set misses
- A paper and its blog post. Different URLs, same result. Cover the paper; mention the post only if it adds an author's clarification.
- Preprint then conference version. Different arXiv id, often a different title, sometimes months apart. If the contribution is the same, it is one item — and the interesting fact is what changed in review.
- The same group's incremental follow-up. "X for Video" three months after "X". Genuinely new, but the listener needs the earlier one as context, so cover them together rather than as two unrelated items.
- Two groups, one idea, same season. Not duplicates. Say so explicitly — independent convergence is itself a signal that the idea is real.
How to decide
Compare on contribution, not on title or authors:
- What does it claim that nobody claimed before?
- What experiment supports that specific claim?
If two items give the same answer to both, they are one item. If they differ on either, they are two — even from the same lab, even with near-identical titles.
When you merge, say so
Never silently drop the second one. A merged entry states it:
Also appeared as a NeurIPS submission with an extra ablation; same core result, so covered once here.
Silent dedup is indistinguishable from a harvest that missed something, and the whole point of a standing sweep is knowing which of the two happened.