Skip to content

fix: harden skill-first research workflow#51

Merged
freshtechbro merged 5 commits into
mainfrom
codex/investigate-search-engine-research-model
May 10, 2026
Merged

fix: harden skill-first research workflow#51
freshtechbro merged 5 commits into
mainfrom
codex/investigate-search-engine-research-model

Conversation

@freshtechbro

Copy link
Copy Markdown
Owner

Summary

  • Harden research search discovery so shell/login/privacy/legal/dead-end pages are rejected as evidence.
  • Keep browser/cookie recovery at destination-level evidence paths while preserving generic provider fallback behavior.
  • Expand research skill outputs and report artifacts with search direction, candidate triage, rejected candidates, deep dives, synthesis feedback, and pending evidence review.
  • Add regression coverage for traversal URL filtering, community follow-up fetch scheduling, renderer artifacts, blocker classification, and evidence bundles.

Validation

  • ./skills/opendevbrowser-research/scripts/validate-skill-assets.sh
  • ./skills/opendevbrowser-best-practices/scripts/validate-skill-assets.sh
  • git diff --check
  • npm run lint
  • npm run typecheck
  • npm run build
  • npm run test (4243 passed, 1 skipped; branch coverage 97%)

Review

  • RepoPrompt final follow-up review: no findings.

Notes

  • Real proof artifact from the final isolated run: .opendevbrowser/research-guided-real-task-final-20260510/output/research/63b24c24-5c55-4421-97de-9298dd751b1c
  • Private mirrored website sync is still a downstream follow-up outside this repo.

freshtechbro and others added 5 commits May 9, 2026 12:57
- Document provider reliability limits and search-engine research tradeoffs

- Capture free SERP API findings and the skill-first implementation plan

- Include plan critique that shaped the implementation guardrails

Co-authored-by: Codex <noreply@openai.com>
- Reframe research as skill-guided and provider-constrained

- Add search-engine discovery as a skill-only, discovery-only lane

- Harden research skill validation with contract-drift regression tests

Co-authored-by: Codex <noreply@openai.com>
- Replace reliable-auto research copy with evidence-gated guidance

- Regenerate public-surface snapshots and update help/onboarding tests

- Align handoff and live-matrix proof commands with explicit source families

Co-authored-by: Codex <noreply@openai.com>
- Detect auth, token, and challenge-gated research failures in handoff metadata
- Suggest signed-in extension recovery and cookies only when diagnostics justify it
- Document user-authorized gated-provider recovery in the research skill

Co-authored-by: Codex <noreply@openai.com>
- reject search, login, privacy, legal, and no-link shells as research evidence
- preserve browser recovery only for destination-level evidence paths
- enrich research artifacts with triage, deep-dive, and synthesis sections
- add regression coverage for guided traversal and honest artifact reporting

Co-authored-by: Codex <noreply@openai.com>
@freshtechbro freshtechbro merged commit 41796b3 into main May 10, 2026
8 checks passed
@freshtechbro freshtechbro deleted the codex/investigate-search-engine-research-model branch May 10, 2026 16:05
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.

1 participant