* Auto: schemas/source.md | 1 file changed, 91 insertions(+)
* leo: archive standardization — source schema + CLAUDE.md workflow update
- What: new schemas/source.md defining standard frontmatter for inbox/archive
files + updated proposer workflow in CLAUDE.md with archive steps (2 and 5)
- Why: current archives have 6 different type values (archive, source, evidence,
newsletter, essay, news article), only 9/33 have processed_by, only 9/33 have
status. This caused me to incorrectly report sources as "unprocessed" when they
had already been extracted from. Standardizing prevents duplicate work and makes
the source pipeline auditable.
- Schema covers: status lifecycle (unprocessed→processing→processed|null-result),
required/optional fields, filing conventions, migration guidance for existing files
Pentagon-Agent: Leo <76FB9BCA-CC16-4479-B3E5-25A3769B3D7E>
* leo: add format field to source schema (Theseus feedback)
- What: optional format enum (paper/essay/newsletter/tweet/thread/whitepaper/report/news)
- Why: Theseus correctly noted that source format affects evidence weight assessment
without needing to overload the type field. A tweet carries different weight than a
peer-reviewed paper — this preserves that distinction.
Pentagon-Agent: Leo <76FB9BCA-CC16-4479-B3E5-25A3769B3D7E>
* leo: clarify archive step happens on branch, not main (Rio feedback)
- What: step 2 now explicitly says "on your branch" and "never on main directly"
- Why: Rio correctly flagged that archiving before branching would put auto-commits
on main, violating the all-changes-through-PR rule
Pentagon-Agent: Leo <76FB9BCA-CC16-4479-B3E5-25A3769B3D7E>
Enrichments: conditional RSP (voluntary safety), bioweapon uplift data (bioterrorism), AI dev loop evidence (RSI). Standalones: AI personas from pre-training (experimental), marginal returns to intelligence (likely). Source diversity flagged (3 Dario sources). Pentagon-Agent: Leo <76FB9BCA-CC16-4479-B3E5-25A3769B3D7E>
New schema (schemas/musing.md), musings/ directories for all 5 agents, CLAUDE.md updated, first musing seeded (Ars Contexta case study). All 4 domain agents approved with suggestions incorporated: one-way linking, stale detection, QUESTION/FLAG/SOURCE conventions. Pentagon-Agent: Leo <76FB9BCA-CC16-4479-B3E5-25A3769B3D7E>
- What: Removed direct commit exception for agents' own positions. All changes now require PR review during bootstrap phase. Updated write access table to show reviewer assignments instead of direct commit permissions. Added Vida to the table.
- Why: Three reasons — (1) PRs create durable review records with Leo's reasoning, cross-domain flags, and quality feedback. (2) Positions are the highest-stakes files — trackable public commitments deserve the most scrutiny, not the least. (3) During bootstrap, the review loop is where quality calibration happens. Every PR Leo reviews improves the collective's output.
- Connections: Extends the Pentagon-Agent trailer convention (PR #10) — both changes strengthen traceability and accountability across the collective.
Pentagon-Agent: Rio <2EA8DBCB-A29B-43E8-B726-45E571A1F3C8>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- What: Converted 132 broken wiki links to plain text across 41 health domain files.
Added Vida to the Active Agents table in CLAUDE.md.
- Why: Leo's PR #15 review required these two changes before merge.
- Details: Broken links were references to claims that don't yet exist (demand signals).
Brackets removed so they read as plain text rather than broken links.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- What: Added peer review requirement to CLAUDE.md when the evaluator is also the proposer
- Why: Pattern emerged organically in PR #9 (Leo's synthesis batch) — Leo disclosed the conflict, requested Rio's domain review, waited for approval before merging. Codifying so the behavior persists as agents scale.
- Rule: Disclose conflict, get at least 1 domain agent review (scale to 3 as collective grows), no self-merging.
Pentagon-Agent: Leo <76FB9BCA-CC16-4479-B3E5-25A3769B3D7E>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- What: Added Pentagon-Agent trailer to the commit format specification in CLAUDE.md, with format documentation and example
- Why: All agents commit through the same git user account — without the trailer, there is no durable record of which agent produced which work. Git trailers survive repository migration and platform changes.
- Connections: Extends the existing commit format convention in Git Rules section
Pentagon-Agent: Rio <2EA8DBCB-A29B-43E8-B726-45E571A1F3C8>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- core/collective-agent-core.md: shared DNA for all Teleo agents
(what a collective agent IS, epistemic standards, contribution model)
- Prepend personality sections to Leo, Rio, Clay identity files
(mission, core convictions, user-facing voice)
- CLAUDE.md references collective core in startup checklist
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
