Skip to content

Harden agents: risk-tiered autopilot, dedup spec, grounding tests - #368

Merged
telivity-otaip merged 4 commits into
mainfrom
cursor/typesafe-harden-agents-7e82
Sep 24, 2026
Merged

telivity-otaip merged 4 commits into
mainfrom
cursor/typesafe-harden-agents-7e82

Conversation

@telivity-otaip

@telivity-otaip telivity-otaip commented Sep 24, 2026 •

Copy link
Copy Markdown
Collaborator

Summary

Hardens HAIP agents without putting TypeSafe (or any third-party AI SDK) into the public repo.

  1. Autopilot risk tiers — guest_comms / review_response never auto-execute; money-moving agents require max(configThreshold, 0.92); ops keep the configured threshold.
  2. Property dedup spec — 4-2 + domain-4 doc: similarity only shortlists; final outcome is same / maybe (human) / different. No merge module.
  3. LLM grounding tests — extra false-clear/false-suppress regressions ($12, 100↛1%, suggestion filtering).
  4. README test counts — synced to 2407 / 281 after the new specs (fixes CI stale-count gate).

Not in this PR: TypeSafe-backed review meaning. That stays out of public haip (Cloud-only if pursued later). Keyword review path unchanged.

Test plan

  • agent-autopilot-tiers + agent.service unit tests
  • grounding.spec.ts (16 tests)
  • Spec/doc review of same/maybe/different wording
  • README / docs/test-stats.json synced to CI actual counts (2407 / 281)
Open in Web Open in Cursor 

cursoragent and others added 3 commits September 24, 2026 06:02
Guest-facing agents never auto-execute. Money-moving agents require
max(configThreshold, 0.92). Ops agents keep the configured threshold.

Co-authored-by: telivity-otaip <telivity-otaip@users.noreply.github.com>
Similarity scores only shortlist candidates. Final merge authority is
three-way: same, maybe (human + field disagreements), or different.

Co-authored-by: telivity-otaip <telivity-otaip@users.noreply.github.com>
Add false-clear edges for small $-amounts, inverse percent scaling, and
suggestion filtering. Align domain-4 lodging doc with same/maybe/different.

Co-authored-by: telivity-otaip <telivity-otaip@users.noreply.github.com>
@cursor cursor Bot changed the title Harden agents: risk-tiered autopilot, TypeSafe reviews, dedup spec, grounding tests Harden agents: risk-tiered autopilot, dedup spec, grounding tests Sep 24, 2026
@telivity-otaip
telivity-otaip marked this pull request as ready for review September 24, 2026 06:06
CI reported 2407 passed tests across 281 files; update README badge
and docs/test-stats.json so the stale-count gate passes.

Co-authored-by: telivity-otaip <telivity-otaip@users.noreply.github.com>
@telivity-otaip
telivity-otaip merged commit d1c463c into main Sep 24, 2026
5 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants