fix: Bind restored review sessions to trusted base

[idy]

Summary

Closes #41.

• Bind the restored model session to the current trusted base and session ID; retain deterministic review checkpoints, but reset stale conversation context after a base change or legacy unbound artifact.
• Supply the current repository/revision on every turn and fail before model invocation if the Git commit is genuinely unavailable.
• Document the lifecycle and exercise runner behavior with Git fixtures and a fake model executable.

Validation

• node .github/scripts/pr-review/test.mjs passed, including same-base resume, changed/missing binding restart, within-run resume and unavailable commit rejection.
• node .github/scripts/issue-review/test.mjs passed.
• node .github/scripts/pr-readiness/test.mjs passed.
• node .github/scripts/review-request/test.mjs passed.
• git diff --check passed.

Evidence and release

In GizClaw/gizos run 34753915398, Actions successfully checked out 1f7f5a3e. The restored Codex session instead executed git ls-tree against historical d48cfd97, then mislabeled the resulting error as a missing new base. No checkout permission or untrusted Git-object access is relaxed.

After merge, update the authorized latest tag and request a fresh GizClaw/gizos#178 review to verify the consuming workflow. This workflow-only fix does not claim firmware acceptance or authorize merging that firmware PR.

[idy]

@codex review

[idy]

Self-review added regression coverage preserving prior incremental findings when the model conversation is reset. All four script test suites pass locally.

@codex review

[github-actions]

✅ OpenAI PR Review: PASS

Conclusion: Ready from the OpenAI review perspective. PR format, linked Issue design, and code/plan conformance passed with no actionable findings.

Review checks

Check	Result
PR format	✅ PASS
Issue design	✅ PASS
Code & plan conformance	✅ PASS

Scope: f64139f938 · af31372b68..f64139f938 · full · 1 diff chunk

Usage: 58s · 394,860 tokens · 80.3% cache hit · 8.465 credits

Summary

All completed chunk reviews are consistent with Issue #41 and contain no actionable findings or plan-conformance blockers.

Review metadata

• Commit: f64139f938
• Range: af31372b68..f64139f938
• Mode: full
• Diff chunks: 1
• Model: gpt-5.6-terra
• Reasoning effort: medium
• Session: repo-1309321116-pr-42-v2
• Generation: 7a514943327fa9fd009a0f891d81c601c20ab5e92b8f71c792d3b0b9bc8085c3
• Evidence: a55f70685ddc7dba4f4c712641a278bb0e78c86cb4df56bede3b3d22a761a4ca

Totals

• Input: 390,321
• Cached input: 313,453
• Cache write: 76,826
• Output: 4,539
• Reasoning: 2,349
• Total: 394,860
• Estimated credits: 8.465
• Credit rate per 1M tokens: 62.5 input / 6.25 cached / 375 output

Token and cache details

Stage	Mode	Target	Time	Input	Cached	Hit	Output	Total	Credits
pr	deterministic	format rules	0s	0	0	0.0%	0	0	0.000
pr	full	pr	7s	19,902	9,533	47.9%	335	20,237	0.833
issue	full	Issue #41	15s	82,295	67,498	82.0%	1,109	83,404	1.763
code	full	chunk 1/1	31s	187,276	139,227	74.3%	2,663	189,939	4.872
code	full	aggregate	5s	100,848	97,195	96.4%	432	101,280	0.998