feat(verify): VAS-2 — gecko verify-docs (check docs claims against reality) - #215
Merged
Conversation
…erdicts (VAS-2)
verify_docs(client, mode) walks every tool on a surface, replays it through the
shipped call path (client.call -> caller/validator, so the auth-host pin +
recorded scrub + SSRF guard all still apply — no new network path), lifts the
outcome with VAS-1's verdict_from_replay, attaches the VerifyVerdict to the op's
graph node (Node.verify, a separate axis from provenance), and returns a
control-plane report {op_id: {status, basis, provenance}} + {verified, refuted,
unverified} counts.
Recorded (default) is the offline-falsifiable baseline: no wire evidence ->
every op honestly UNVERIFIED, a recorded 200 never overclaimed. Live -> real
VERIFIED / REFUTED; a 404 on a doc-claimed endpoint REFUTES (the fabricated-
endpoint case). op_provenance (the claimed-until-checked classifier) lives in
graph.py so the provenance ladder keeps one source of truth.
CLI: gecko verify-docs <spec> [--live] — thin transport, default recorded.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Milestone-1 step 2.
gecko verify-docs <spec> [--live]checks a surface's operations against reality and attaches VAS-1 verdicts — the Privy-fabrication catch, generalized.What's here —
gecko/verify.pyverify_docs(client, *, mode="recorded")walks every tool, replays it through the already-shipped call path (client.call→ caller/validator — auth-host pin + SSRF guard + recorded scrub all still apply, no new network path), lifts each outcome with VAS-1'sverdict_from_replay, attaches theVerifyVerdictto the op node, and returns a control-plane report{op: {status, basis, provenance}}+{verified, refuted, unverified}.no-access:recorded-only). The offline-falsifiable baseline.--live): real replay → 2xx/3xx VERIFIED, 404/4xx-5xx REFUTED (the fabricated endpoint). Reads the actual returned mode, so a live→recorded degrade stays UNVERIFIED, never falsely verified.Provenance as a separate axis
op_provenance(spec, op)(ingraph.py, CLAIMED's single source of truth) → CLAIMED for a from-docs draft, else EXTRACTED.verify_docsnever overwrites provenance — an op keeps its provenance and gains a verdict.Safety
Never overclaims (UNVERIFIED unless wire-observed — VAS-1's lift enforces it, tested); control-plane clean (report is status/basis/provenance + counts, no response value — asserted); reuses the gated network path.
report.pyuntouched (Scorecard badges are VAS-3).Tests:
tests/test_verify_docs.py(7). Gate: ruff + mypy clean · full suite 1931 passed / 1 skipped.Next: VAS-3 (Scorecard renders CLAIMED/VERIFIED/REFUTED), VAS-5 (the recorded Privy demo).
🤖 Generated with Claude Code