This repository is part of CompText.
CompText is a deterministic Context Pack / proposal-gated evidence workflow, not a generic AI dashboard, not an agent memory product, and not an autonomous deployment agent.
Core line: Models are providers. Context is the product.
Pipeline: Source / GitHub URL → Goal → Inspect → Context Pack → Policy Gate → Provider Boundary → Untrusted Proposal → Human Review → Artifacts
Hard rules:
- Treat GitHub as read-only unless the human explicitly asks otherwise.
- Do not push, deploy, create PRs, create issues, or create remote branches.
- Do not expose secrets.
- Do not fake hashes.
- Do not claim production-ready, EU AI Act compliance, legal certification, forensic proof, guaranteed correctness, or autonomous approval.
- Provider output is untrusted until reviewed.
- Proposals are never auto-applied.
- Human review is the approval boundary.
- Artifacts preserve the evidence trail.
- Keep changes scoped and reviewable.
- Do not modify root
README.mdunless the human explicitly approves it. - Do not commit
reports/latest.json. - Do not commit
reports/performance_baseline.jsonwhen it is only validation churn. - Run cargo commands only inside
agy7rust/unless the human explicitly approves otherwise.
Codex Desktop governance:
- Repo-local hooks live under
.codex/and must be reviewed/trusted by Codex before enforcement. - Hooks block unauthorized git writes, deploy/release actions, environment dumps, secret-file reads,
agy-ct run, andagy-ct benchmark. - Hooks warn on protected source, README, report, and
artifacts/spark/changes; warnings do not make generated artifacts commit-ready.
Before editing:
- Read AGENTS.md.
- Read
.agents/skills/**/SKILL.mdrelevant to the task; treat.agent/skills/*.mdas legacy/compatibility metadata only if present. - If
.agents/skills/00_project_system/SKILL.mdexists, read it too. - Build a compact repo map.
- Stop for approval if the user asked for plan mode.