Lead README with the security findings and supply-chain framing

[jothimani-rajendran]

What

README-only refresh. The README now opens with security: context-report as supply-chain trust for agent context, meaning a signed, re-derivable report of whether a plugin, hook, skill, AGENTS.md or MCP server actually works.

• Header: a new tagline, plus a second row of brand badges for in-toto Statement/v1, Sigstore via actions/attest, 11 report rows and the v0.1 draft spec. All existing badges are kept.
• "Why it matters": a table that pairs each measured finding with what it means for security:
  • 3 of 18 public plugins are reachable but not executable (exit 126).
  • Every hook that runs allows on malformed input, so it fails open.
  • Latency and context-token costs vary by two orders of magnitude.
  • No efficacy row reaches PASSED.
• Signing: a paragraph taken from the spec's Envelope section covering DSSE, Sigstore, Fulcio, Rekor and gh attestation verify. It says producers SHOULD sign; it does not claim the CLI signs.
• "What a report proves": a table covering all 11 rows. For each row it gives the question, why it matters for security, and its honest v0.1 producer status (measured, claimed or NotAvailable).
• Family table: now has column headers. chock-catalog figures are updated to 48 policies (19 enforced-at-commit, 9 in-agent, 20 advisory), 1,179 eval cases and 1,019 replayed in CI. The generated family-*.svg figure still says 39; it was not touched because figures are script-generated.
• Contribute: a table of the first contributions, each with the file it lives in.

The quickstart section is byte-identical, so docs/quickstart.sh does not drift. Every existing figure, number and caveat is kept.

Definition of done

• python -m ruff check . and python -m ruff format --check . clean
• python -m pytest -q green (407 passed, 1 skipped)
• Schema changes: n/a
• chock check and chock sync --repo . --check clean (pinned chock v0.9.0)

Also checked:

• The quickstart.yml drift check matches.
• The quickstart block runs in a fresh directory and ends with well-formed and bound: True.
• The figures job regenerated all figures with no diff.

Claims

• No row is described as re-derivable unless the reference producer measures it. Rows that are not measured are shown as NotAvailable, and efficacy is shown as claimed.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LAxde3s5KW8N36kS8LF3er

---

Generated by Claude Code