clawmates/migrations/0047_missions.sql

-- Slice 1 of the missions unification (see design notes).
--
-- A `mission` is a single top-level user-driven workflow. It replaces
-- the current two-headed split between `research_topics` and `loops`:
-- both of those become special cases of a mission composed of phases.
--
-- This migration ONLY adds the new tables + backfills existing rows.
-- Old tables (research_topics, loops, research_topic_agents,
-- research_outcomes, ...) stay live and writable until Slice 9's
-- big-bang cutover deletes them. Reads route through the compat
-- shims added in Slice 2; the old wizards keep working unchanged.
--
-- Discriminators — kept as TEXT (not enums) so new templates + phase
-- kinds ship as PRs without schema migrations.
--   missions.template_kind ∈
--     {research_only, research_and_code, security_hardening,
--      refactor, benchmark, custom}
--   mission_phases.kind ∈
--     {research, coding, benchmark, security_scan}
--   mission_tasks.status ∈
--     {created, working, validating, complete, failed}
--   mission_artifacts.kind ∈
--     {md, pdf, benchmark_result, security_report, code_diff, index}

CREATE TABLE missions (
    id                UUID PRIMARY KEY,
    workspace_id      UUID NOT NULL,
    title             TEXT NOT NULL,
    template_kind     TEXT NOT NULL,
    -- team_id is nullable so a mission can start without a bound team
    -- (auto-provision materializes one during phase startup).
    team_id           UUID REFERENCES teams(id) ON DELETE SET NULL,
    -- team_template_id captures template lineage — level-up promotions
    -- diff against the template version the team was minted from.
    team_template_id  UUID,
    repo_id           UUID REFERENCES repos(id) ON DELETE SET NULL,
    -- schedule JSONB carries the trigger config (cron | one_shot |
    -- on_event). Kept as JSONB so we can grow the schedule surface
    -- without a table alter each time.
    schedule          JSONB NOT NULL DEFAULT '{"kind":"one_shot"}'::jsonb,
    -- Mission-level status. Rolls up phase statuses per lifecycle rules
    -- enforced in cm-api::routes::missions.
    -- draft → running → (completed | failed | cancelled)
    status            TEXT NOT NULL DEFAULT 'draft',
    -- Human-facing description + optional target subject. Both are
    -- shown in the canvas overview and passed to the LLM as context.
    description       TEXT,
    -- Free-form config for the specific workflow template (task
    -- template overrides, benchmark commands, LLM overrides for PDF
    -- rendering, etc.). Every template contributes its own keys.
    config            JSONB NOT NULL DEFAULT '{}'::jsonb,
    created_at        TIMESTAMPTZ NOT NULL DEFAULT now(),
    updated_at        TIMESTAMPTZ NOT NULL DEFAULT now(),
    completed_at      TIMESTAMPTZ
);
CREATE INDEX missions_workspace_idx
    ON missions (workspace_id, created_at DESC);
CREATE INDEX missions_team_idx
    ON missions (team_id) WHERE team_id IS NOT NULL;
CREATE INDEX missions_status_idx
    ON missions (status, updated_at DESC);

-- A phase is one segment of a mission's plan. Ordering by order_idx.
-- Multiple phases of the same kind are allowed (e.g. two coding
-- passes bracketing a benchmark phase).
CREATE TABLE mission_phases (
    id                UUID PRIMARY KEY,
    mission_id        UUID NOT NULL REFERENCES missions(id) ON DELETE CASCADE,
    kind              TEXT NOT NULL,
    order_idx         INT  NOT NULL,
    status            TEXT NOT NULL DEFAULT 'pending',
    -- Phase-scoped config (e.g. commit_policy, loop bounds, benchmark
    -- cmd) merged over the template's phase spec at run time.
    config            JSONB NOT NULL DEFAULT '{}'::jsonb,
    started_at        TIMESTAMPTZ,
    completed_at      TIMESTAMPTZ,
    UNIQUE (mission_id, order_idx)
);
CREATE INDEX mission_phases_mission_idx
    ON mission_phases (mission_id, order_idx);
CREATE INDEX mission_phases_status_idx
    ON mission_phases (status) WHERE status IN ('pending', 'running');

-- A task is one addressable unit inside a phase. For coding phases
-- that consume an integrations artifact, tasks correspond 1:1 with
-- INT-XX items parsed from run output (Slice 5).
-- Research phases materialize one task per "outcome iteration".
-- Security phases materialize one task per finding.
CREATE TABLE mission_tasks (
    id                UUID PRIMARY KEY,
    mission_id        UUID NOT NULL REFERENCES missions(id) ON DELETE CASCADE,
    phase_id          UUID NOT NULL REFERENCES mission_phases(id) ON DELETE CASCADE,
    -- external_id: e.g. "INT-05", "CVE-2024-1234", "OUTCOME-v3".
    -- Nullable so ad-hoc tasks work without a parseable marker.
    external_id       TEXT,
    title             TEXT NOT NULL,
    -- Assigned agent (nullable — some tasks are team-wide).
    assigned_agent_id UUID REFERENCES agents(id) ON DELETE SET NULL,
    status            TEXT NOT NULL DEFAULT 'created',
    -- The topology run that produced this task's most recent state
    -- update. Follows the run's lifecycle for observability.
    run_id            UUID REFERENCES topology_runs(id) ON DELETE SET NULL,
    -- Paths to the artifacts this task produced (relative to the
    -- mission's artifact root). One task can have many artifacts;
    -- kept as an array for cheap lookup without a join table.
    artifact_paths    TEXT[] NOT NULL DEFAULT '{}',
    created_at        TIMESTAMPTZ NOT NULL DEFAULT now(),
    updated_at        TIMESTAMPTZ NOT NULL DEFAULT now(),
    completed_at      TIMESTAMPTZ
);
CREATE INDEX mission_tasks_mission_idx
    ON mission_tasks (mission_id, created_at);
CREATE INDEX mission_tasks_phase_idx
    ON mission_tasks (phase_id, status);
CREATE INDEX mission_tasks_agent_idx
    ON mission_tasks (assigned_agent_id) WHERE assigned_agent_id IS NOT NULL;
-- Uniqueness on (phase_id, external_id) so the task-card parser (Slice
-- 5) can UPSERT on marker match without duplicating rows.
CREATE UNIQUE INDEX mission_tasks_external_uniq
    ON mission_tasks (phase_id, external_id)
    WHERE external_id IS NOT NULL;

-- Artifacts land on the filesystem under a per-mission root
-- (see cm-api::missions::artifact_root). We index them here for
-- typed discovery from the UI without walking the filesystem.
CREATE TABLE mission_artifacts (
    id                    UUID PRIMARY KEY,
    mission_id            UUID NOT NULL REFERENCES missions(id) ON DELETE CASCADE,
    phase_id              UUID REFERENCES mission_phases(id) ON DELETE SET NULL,
    -- Repo-relative path from the mission root, e.g.
    -- "research/v3/spec.md" or "benchmarks/before.json".
    path                  TEXT NOT NULL,
    kind                  TEXT NOT NULL,
    mime                  TEXT,
    -- Optional back-pointer to the run that produced this artifact.
    generated_by_run      UUID REFERENCES topology_runs(id) ON DELETE SET NULL,
    -- Sidecar for the PDF renderer (Slice 6). NULL until rendered.
    rendered_pdf_path     TEXT,
    render_pdf_status     TEXT NOT NULL DEFAULT 'skip',
    -- ↑ skip | pending | rendering | done | failed
    render_pdf_error      TEXT,
    -- Human-shown title; falls back to filename when NULL.
    title                 TEXT,
    -- Free-form metadata (word count, sha256, source model, etc.).
    metadata              JSONB NOT NULL DEFAULT '{}'::jsonb,
    created_at            TIMESTAMPTZ NOT NULL DEFAULT now(),
    updated_at            TIMESTAMPTZ NOT NULL DEFAULT now(),
    UNIQUE (mission_id, path)
);
CREATE INDEX mission_artifacts_mission_idx
    ON mission_artifacts (mission_id, created_at DESC);
-- PDF renderer picks up its queue via this partial index.
CREATE INDEX mission_artifacts_pdf_queue_idx
    ON mission_artifacts (created_at)
    WHERE render_pdf_status = 'pending';

-- Benchmark snapshots — one row per benchmark phase per iteration.
-- Stored as JSONB pairs so the shape can evolve per template
-- (criterion, cargo bench, k6, wrk, custom scripts) without alters.
CREATE TABLE benchmark_snapshots (
    id                UUID PRIMARY KEY,
    mission_id        UUID NOT NULL REFERENCES missions(id) ON DELETE CASCADE,
    phase_id          UUID NOT NULL REFERENCES mission_phases(id) ON DELETE CASCADE,
    -- iteration = 0 for the pre-coding baseline; ≥1 for post-iteration
    -- snapshots. Uniqueness enforced so re-runs overwrite in place.
    iteration         INT  NOT NULL,
    before_metrics    JSONB,
    after_metrics     JSONB,
    delta             JSONB,
    -- Optional descriptor of the benchmark command / driver.
    driver            TEXT,
    created_at        TIMESTAMPTZ NOT NULL DEFAULT now(),
    UNIQUE (phase_id, iteration)
);
CREATE INDEX benchmark_snapshots_mission_idx
    ON benchmark_snapshots (mission_id, created_at DESC);

-- ── Backfill ─────────────────────────────────────────────────────
--
-- Every existing research_topic becomes a mission with a single
-- research phase. Every existing loop with source_research_topic_id
-- becomes a research_and_code mission with two phases pointing at the
-- same underlying topic + loop rows through the compat shim.
--
-- Standalone loops (no source topic) become refactor missions with a
-- single coding phase.
--
-- This is a one-shot copy — old rows are NOT deleted (Slice 9). The
-- compat shims read from `missions` first, fall through to the
-- legacy tables when nothing landed, so both surfaces stay coherent
-- during the transition.

-- 1. Research-only backfill.
INSERT INTO missions (
    id, workspace_id, title, template_kind, team_id, repo_id,
    schedule, status, description, config, created_at, updated_at
)
SELECT
    id,
    workspace_id,
    title,
    'research_only',
    team_id,
    repo_id,
    '{"kind":"one_shot"}'::jsonb,
    CASE
        WHEN status IN ('publishing','complete','archived') THEN 'completed'
        WHEN status = 'processing'                          THEN 'running'
        WHEN status = 'failed'                              THEN 'failed'
        ELSE 'draft'
    END,
    description,
    jsonb_build_object(
        'legacy_topic_id', id::text,
        'outcome_kind',    outcome_kind,
        'topology_kind',   topology_kind
    ),
    created_at,
    updated_at
FROM research_topics
WHERE NOT EXISTS (SELECT 1 FROM missions m WHERE m.id = research_topics.id);

INSERT INTO mission_phases (id, mission_id, kind, order_idx, status, config)
SELECT
    gen_random_uuid(),
    id,
    'research',
    0,
    CASE
        WHEN status IN ('publishing','complete','archived') THEN 'completed'
        WHEN status = 'processing'                          THEN 'running'
        WHEN status = 'failed'                              THEN 'failed'
        ELSE 'pending'
    END,
    jsonb_build_object('legacy_topic_id', id::text)
FROM research_topics
WHERE NOT EXISTS (
    SELECT 1 FROM mission_phases mp
    WHERE mp.mission_id = research_topics.id AND mp.kind = 'research'
);

-- 2. Loop backfill — one mission per loop, template_kind chosen by
-- whether it's paired to a research topic.
INSERT INTO missions (
    id, workspace_id, title, template_kind, team_id, repo_id,
    schedule, status, description, config, created_at, updated_at
)
SELECT
    l.id,
    l.workspace_id,
    l.title,
    CASE
        WHEN l.source_research_topic_id IS NOT NULL THEN 'research_and_code'
        ELSE 'refactor'
    END,
    l.team_id,
    NULL,
    jsonb_build_object(
        'kind',     CASE WHEN l.enabled THEN 'cron' ELSE 'one_shot' END,
        'triggers', l.triggers
    ),
    'running',
    l.description,
    jsonb_build_object(
        'legacy_loop_id',        l.id::text,
        'source_research_topic', l.source_research_topic_id::text,
        'loop_kind',             l.kind
    ),
    l.created_at,
    l.updated_at
FROM loops l
WHERE NOT EXISTS (SELECT 1 FROM missions m WHERE m.id = l.id);

-- 2a. Research-then-code loops get two phases (research phase already
-- exists as the paired topic mission; we add a coding phase to this
-- mission that points at the same loop_id via config).
INSERT INTO mission_phases (id, mission_id, kind, order_idx, status, config)
SELECT
    gen_random_uuid(),
    l.id,
    'coding',
    0,
    'pending',
    jsonb_build_object(
        'legacy_loop_id',              l.id::text,
        'source_research_topic_id',    l.source_research_topic_id::text
    )
FROM loops l
WHERE NOT EXISTS (
    SELECT 1 FROM mission_phases mp
    WHERE mp.mission_id = l.id AND mp.kind = 'coding'
);