fix(agentic-engineering): bind autonomy recommendations and retain known loss

[devantler]

🤖 Generated by the Agentic Engineer

Why

An autonomy assessment could recommend from a foreign protected request, reuse one report across evidence stages, or lose a known runtime failure when the counterpart report became incomplete. Those results could mislead an operator deciding whether to retain or contract a candidate.

What

Require distinct stage records, bind protected requests before retaining the incumbent, and assess each bound runtime failure independently. Complete positive and intercepted-negative reports remain required for candidate support. Ship the fixes in agentic-engineering 6.1.1 with updated consumer guidance.

Validation: assessor RED/GREEN cases, existing Go coverage, the copied installed command, full ShellCheck inventory, manifests and version/changelog gates pass. The command remains advisory and never runs recovery or admits execution.

Fixes #465
Fixes #466
Fixes #467

[devantler]

🤖 Generated by the Agentic Engineer

Self-review (fallback — CodeRabbit, Codex and Cursor Bugbot unavailable)

Reviewed commit: a61dea3

• CodeRabbit: no artifact exists on this PR at the reviewed head; the account-wide rate-limit refusal was observed at 2026-10-04T16:32:24Z on devantler-tech/platform#4474 (comment) and resets at 2026-10-04T17:09:22Z.
• Codex: no artifact exists on this PR at the reviewed head; the account-wide usage-limit refusal was observed at 2026-10-04T02:11:02Z on devantler-tech/ksail#7433 (comment) and requires account credits or settings recovery.
• Cursor Bugbot: no artifact or check exists on this PR at the reviewed head; the user/team usage-limit refusal was observed at 2026-10-04T12:21:20Z on devantler-tech/ksail#7481 (comment) and requires usage or spend recovery.
• Direct current-PR review objects, comments, threads, and Bugbot checks were read before this fallback; no provider finding is being discarded.

Verdict: no P0/P1 findings

[devantler]

🤖 Generated by the Agentic Engineer

Self-review (fallback — CodeRabbit, Codex and Cursor Bugbot unavailable)

Reviewed commit: a61dea3

• CodeRabbit: freshly verified authenticated review 5407216028, event 2026-10-04T16:45:47Z, states one included review per hour and zero remaining. This is the applicable included account allowance; recovery requires its renewal or a newer serving artifact. No overage was requested.
• Codex: freshly verified authenticated account limit, event 2026-10-04T02:11:02Z; account-wide review usage exhausted, no reset specified. Recovery requires included quota renewal. Paid credits were not enabled.
• Cursor Bugbot: freshly verified authenticated user/team limit, event 2026-10-03T22:35:03Z; usage/spend limit applies to this user or team, no reset specified. Recovery requires included capacity; no spend increase was made.

All current-PR comments, reviews, complete thread inventory and native checks were read before this round. There are no outstanding findings or threads; all 45 current-head checks are settled with CI - Required Checks successful.

Reviewed all eight changed files and the complete assessor paths. Protected requests validate scope, immutable bindings and current incumbent before recommending retention. Stage record uniqueness complements ID checks without replacing required-kind coverage. Runtime loss uses each report's own scope, provenance, chronology and path coverage; candidate support still needs distinct complete positive and intercepted-negative reports. Foreign and future failures remain HOLD. Scoped tested recovery is still required for contraction. The four version manifests agree at 6.1.1; generated-resource hashes are unchanged because no pinned role/runtime asset changed.

Independent read-only review found no correctness issues. The new regressions first failed on the original assessor; the complete Go suite and copied installed consumer pass. As a user, I ran the installed assessment command on the valid synthetic proof and the contraction case, observing RECOMMEND_CANDIDATE and RECOMMEND_CONTRACTION respectively, with executionAdmitted, mutationPerformed and reportedEvidenceAuthenticated false. No report was authenticated, runtime enabled or recovery executed.

Verdict: no P0/P1 findings

[devantler]

🤖 Generated by the Agentic Engineer

Post-merge verification passed from freshly fetched main containing 2a2a35b. The installed assessment consumer and its Go regression suite retain distinct stage evidence, bind protected requests before retaining the incumbent and preserve independently bound runtime failures. The assessment remains advisory. Issues #465, #466 and #467 are closed and their project cards were read back as Done.