Trust without trust-me — immutable prompts & CI guards

Contributors may only edit data/ in PRs. Everything else is protected:

LayerWhat it blocks
contributor-guard.ymlEdits to scripts, agents, CI from report branches
validate-report.ymlBad schema, fake URLs, missing frontmatter
prompt_hash in reportsProves which template version was used
ci-integrity.ymlCloud anchor — guard scripts can't be silently removed

Prompt provenance


prompt_hash: 16b431faca503ced
prompt_file: agents/investigation-social.md

Validator recomputes hash from repo template. Tamper-evident research.

Multi-report integrity

One file per user per ticker per day:


report.alice.md   ← Alice can't overwrite Bob's file
report.bob.md

Roadmap anti-manipulation

Open source means auditable rules, not "trust our black box."


TRUST.md

Series: Market AI on Git · #13 of 15

View on GitHub Gist · Star the repo