automation: hypatia ruleset + gitbots should be able to fan-out doc + 6a2 + contractile refreshes themselves #348

Origin

2026-06-02 wave: Claude orchestrated 25+ PRs across echidna+echidnabot+gitbot-fleet+hyperpolymath doing doc maxout, 6a2 STATE refreshes, contractile family completion (~1500 LoC across both repos), saturation+typing wiring, training run, license reconciliation, branch + /tmp cleanup, and assorted bug fixes (JSON dep, CUDA.version(), shared-context edition).

The doc/6a2/contractile work in particular was mechanical-but-large: each repo needed:

• README + EXPLAINME currency sweep (catch stale claims, link to recent PRs, drop sunset-era figures)
• .machine_readable/6a2/STATE.a2ml last-updated + session-summary refresh
• contractiles/{must,adjust,trust,bust,dust}/* fill-in (rsr-template-repo skeletons → repo-specific content with error codes + recovery cross-references)

Each took multiple turns by Claude inline because the agent fan-out kept hitting the 600s stream watchdog on these mid-size tasks.

Ask

Two automation surfaces could absorb this:

1. Hypatia ruleset additions

• R-DOC-STALE: flag EXPLAINME / 6a2 STATE files whose last-touched git timestamp is > 30 days old AND repo has new merged PRs in that window (suggesting drift).
• R-CONTRACTILE-DRIFT: flag contractile sources still containing rsr-template-repo phrases (e.g. 'project-state-contract', 'config/service.yaml' placeholders) or wrong-repo-context content (e.g. 'Burble' in non-Burble repos).
• R-6A2-MINIMAL: flag 6a2 STATE.a2ml whose [metadata].version doesn't match Cargo.toml/CHANGELOG/git-tag triplet (R5b style but for 6a2).

2. Gitbot fleet recipe

A scheduled gitbot job ('docbot' or extend rhodibot) that:

• Per repo: runs the 3 R-* rules above; if any fire, opens a PR with a STATE.a2ml + EXPLAINME refresh diff for owner review.
• Owner-confirmable so the PR doesn't auto-merge but the work is done.

Why this matters

Otherwise this pattern repeats every major-wave session: hours of mechanical doc-currency work that AI is fine at but agent timeouts make unreliable. A cheap rule-driven nudge plus a gitbot-prepared PR would dramatically lower the operational cost.

Related

• Today's wave snapshot in [[session_2026_06_01_echidna_echidnabot_day]] (Claude memory)
• echidnabot#77 + chore(a2ml): remove 8 deprecated legacy-Scheme example dups (Item 10) #78 + echidna#214 are the contractile family completion examples to model from
• echidnabot#81 + echidna#215 are the 6a2 STATE refresh examples