Skip to content

feat(plan): actionable errors, check lifecycle, soft plan enforcement - #271

Merged
jkyberneees merged 6 commits into
mainfrom
feat/plan-tool-accuracy
Sep 26, 2026
Merged

jkyberneees merged 6 commits into
mainfrom
feat/plan-tool-accuracy

Conversation

@jkyberneees

Copy link
Copy Markdown
Contributor

Summary

  • Actionable validation errors: every plan-tool rejection now carries the failing field path, the violated rule, a retryable flag, and a full example payload; gating errors name the blocking check ids with the exact satisfying tool calls.
  • Check lifecycle: new check_replace verb replaces dead/stale checks (fresh check or equivalent-evidence note, mandatory audit-trailed justification); environment-denied checks transition to blocked which no longer gates step completion (closeout honesty only); create may always reset — the superseded checked plan is archived in the revision block, eliminating deadlock states.
  • Soft runtime enforcement: planning.remind config (default OFF) injects one bounded reminder after 3 plan-less tool calls and flags late plans provisional; never a hard gate. Project config cannot re-enable it when the operator disabled it.
  • Baseline eval harness: 20-scenario scripted eval with failure taxonomy (test-only, CI-hermetic) — first-attempt success 35% → 40%, deadlock class eliminated.
  • Docs: docs/CONFIG.md and docs/PLANNING.md updated in the same change.

Test plan

  • RED-first tests per package; full internal/loop, internal/config, internal/session, internal/telegram suites green; go vet clean; E2E with head binary shows no regressions (sandbox/Docker and API-key failures are pre-existing and environmental).
  • 3 adversarial reviewers over the branch diff; all found defects fixed (single notify per check_replace, rune-safe truncation, wire rollback note).

…le flag, gating check ids with exact satisfying calls (P2)
…, check_replace verb with evidence path, create always resets with archived revision, verb schema surfaced (P3)
…ult OFF), reminder at 3 plan-less calls, provisional late-plan flag; docs updated
…d flag, rune-safe evidence_note truncation, wire rollback note
@cloudflare-workers-and-pages

cloudflare-workers-and-pages Bot commented Sep 26, 2026 •

Copy link
Copy Markdown

Deploying with  Cloudflare Workers  Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status Name Latest Commit Preview URL Updated (UTC)
✅ Deployment successful!
View logs
odek 7cc1524 Commit Preview URL

Branch Preview URL
Sep 26 2026, 03:32 PM

@jkyberneees
jkyberneees merged commit 6706cc7 into main Sep 26, 2026
10 checks passed
@jkyberneees
jkyberneees deleted the feat/plan-tool-accuracy branch September 26, 2026 15:38
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