feat: hotato candidate -- measured candidate-identity binding (honest Candidate Revision proof)#39
Merged
Merged
Conversation
The strongest proof scope, Candidate Revision, previously elevated on an operator-ASSERTED --candidate-config-hash string: nothing verified it against the real config. hotato candidate closes that: - candidate hash --provider vapi --assistant <id> fetches the assistant config, canonicalizes it (strips volatile id/orgId/createdAt/updatedAt recursively, keeps the semantic fields), and prints a deterministic sha256 content hash. - candidate verify --expect <hash> re-fetches after the run and refuses (exit 1) on drift, so a Candidate Revision proof cannot survive a swapped candidate. The hash is a MEASURED binding of the configuration, not runner authentication (evidence_authority stays measured). Pure canonicalize/hash logic is stdlib-only and fully hermetic-testable; the single network fetch reuses capture.py's SSRF-guarded, credential-safe HTTP path. Live call placement stays operator- gated (drive). Verified: volatile noise -> identical hash, semantic change -> different hash; 19 candidate tests, adversarially confirmed. No version bump. Closes the audit's candidate-bound-proof-path engineering gap (hermetic half). Claude-Session: https://claude.ai/code/session_01BqJW5Dey1DzqBeBCAJXrH9
…ety) The candidate verify example used --expect sha256:<hash>; a bare < reads as a shell redirect, failing test_all_help_epilogs_are_paste_safe. Quote it as --expect 'sha256:THE_HASH' like the repo's other paste-safe examples.
hotato turn-taking eval8 of 8 scenarios pass. 0 fail. No regression.
RegressionsNone. Reproducible timing measured locally from call audio. Swap the bundled self-test step for your own captured recordings to gate on your agent. github.com/attenlabs/hotato |
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.
Closes the audit's candidate-bound-proof-path engineering gap (hermetic half).
Problem: hotato prove --candidate-config-hash X --provider vapi claimed the strongest scope (Candidate Revision) on an operator-ASSERTED string; nothing verified it against the real config.
Fix — a measured binding:
candidate hash --provider vapi --assistant <id>: fetch the assistant config, canonicalize (strip volatile id/timestamps, keep semantic fields), print a deterministic sha256 content hash.candidate verify --expect <hash>: re-fetch after the run, refuse (exit 1) on drift, so a Candidate Revision proof cannot survive a swapped candidate.The hash is a measured binding of the configuration, not runner authentication (evidence_authority stays measured). Pure logic is stdlib-only + hermetic; the single network fetch reuses capture.py's SSRF-guarded path. Live call placement stays operator-gated. Verified: volatile noise -> identical hash, semantic change -> different hash; 19 candidate tests, adversarially confirmed. Full gate green (4762). No version bump.
https://claude.ai/code/session_01BqJW5Dey1DzqBeBCAJXrH9