fix(kanban): freeze repeated identical failures

[jkobject]

Summary

• persist a normalized failure fingerprint and freeze a task after its third materially identical terminal outcome
• enforce the freeze across claim, review, reclaim, promotion, ready recomputation, auto-decomposition, dashboard, and tool paths
• require an audited operator/CTO rearm, while keeping punctuation-only diagnostics equivalent and distinct root causes separate
• retain backward compatibility for boards frozen by the earlier block-loop event circuit breaker

Observable behavior

Attempts one and two may be explicitly retried. The third materially identical failure atomically routes the card to triage and makes every automatic launch/reclaim/promotion/decomposition path ineligible. A reasoned operator rearm clears the breaker and restores eligibility.

Test plan

• scripts/run_tests.sh tests/hermes_cli/test_kanban_block_kinds.py tests/hermes_cli/test_kanban_core_functionality.py tests/hermes_cli/test_kanban_db.py tests/hermes_cli/test_kanban_decompose.py tests/plugins/test_kanban_dashboard_plugin.py tests/tools/test_kanban_tools.py — 679 passed
• python3 -m py_compile ... — passed
• git diff --check origin/main...HEAD — passed
• full repository suite running on the combined reconciliation candidate; result will be posted before review handoff

Provenance

This semantically ports the accepted local block-loop chain and the independently reviewed third-identical-failure hardening onto current main, preserving original commit authorship.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[jkobject]

Final reconciliation verification at combined candidate 4e45d6bb870ac9e9e0c01463dd8a9838c92fe19e:

• all changed-scope suites passed (Kanban freeze 679; Supermemory 89; graph policy/prompt/config 464)
• Ruff, py_compile, git diff --check, isolated --version, and isolated doctor passed
• the broad scripts/run_tests.sh run discovered ~40,584 tests; 51 failures occurred in 15 unrelated platform/environment files
• the exact 15-file subset was rerun with -j 1 on both the candidate and clean origin/main@e89bc58a; both produced the identical 51 failures, so the broad-run reds are baseline/environment failures rather than regressions in this PR

No deployment or merge was performed.

[teknium1]

Thanks for the detailed failure-loop hardening and coverage. The premise remains present on current main: block_task equates a repeated cause with only prev_kind == kind and escalates at the two-recurrence threshold (hermes_cli/kanban_db.py:5571-5580), while gateway auto-decomposition processes all triage IDs (gateway/kanban_watchers.py:1367-1381). The durable fingerprint and automatic-path guards address that gap.

I found no verified blocking defect in the proposed implementation. This is an automated hermes-sweeper review.

[GottZ]

This was generated by AI during triage.

Summary

Thirteen PRs address or reference #59333/#66909; #59367, #59536, and #59645 implement the requested global kanban.block_recurrence_limit, while the remaining diffs provide the merged loop-breaker foundation or address distinct review, dependency, authorization, cron, PASS-loop, and repeated-failure paths. Contributor discussion identifies #66909 as a duplicate of #59333 and #59367 as the active implementation.

Related pull requests

• feat(kanban): typed block reasons + unblock-loop breaker #52848 [merged] related — (+551/-49) — n/a: Merged reference implementation that introduced typed block reasons, persistent recurrence counting, and the hard-coded BLOCK_RECURRENCE_LIMIT = 2; it remains relevant as the implementation that created the configuration gap reported here.
• fix(kanban): enforce review change loop before completion #58695 related — (+84/-1) — n/a: Exempts review-required: reblocks and adds a comment-derived completion gate rather than configuring the general limit; retain only the focused reblock exemption unless the contributor keep_open review’s authorized-reviewer and deterministic-ordering objections are resolved.
• feat(kanban): make block recurrence limit configurable #59367 best fix — (+86/-3) — n/a: Adds kanban.block_recurrence_limit to defaults, uses its validated value for routing and event telemetry, documents it, and tests threshold and boolean fallback behavior; this is the recorded best fix for both [Feature]: Use config for BLOCK_RECURANCE_LIMIT #59333 and [Feature]: Please allow to modify BLOCK_RECURRENCE_LIMIT #66909 and addresses its contributor keep_open review.
• feat(kanban): make block recurrence limit configurable #59536 fixes — (+76/-7) — n/a: Implements the same global config key and routing substitution as feat(kanban): make block recurrence limit configurable #59367, but despite its keep_open review the visible diff accepts booleans and truncates non-integral floats, making it a weaker duplicate of feat(kanban): make block recurrence limit configurable #59367.
• feat(cron): make BLOCK_RECURANCE_LIMIT configurable (#59333) #59645 partial — (+415/-2) — n/a: Reimplements the global override with extensive tests and example configuration, but despite its keep_open review it omits DEFAULT_CONFIG registration and logs invalid configuration on every resolver call; the contributor discussion identifies feat(kanban): make block recurrence limit configurable #59367 as the canonical duplicate target.
• fix(kanban): guard needs_input unblock from non-interactive sessions #59906 related — (+451/-2) — n/a: Adds a DB/process-ancestry guard and an environment-scrubbed CLI E2E test for worker self-unblocking of needs_input; consistent with the maintainer-bot keep_open verdict, this is a separate authorization safeguard, not recurrence-limit configuration.
• feat(cron): add configurable max_repeat safeguard for cron job repeats (#59333) #60476 related — (+142/-0) — n/a: Adds cron.max_repeat clamping only during cron-job creation and does not alter Kanban’s recurrence threshold; its keep_open review requires re-scoping, centralized create/update validation, default registration, and documentation.
• fix: stop dependency-wait kanban promotion loops #61366 related — (+164/-38) — n/a: Prevents parentless dependency waits from immediate redispatch by routing them to triage while preserving parent-gated waits, and its amended interface, documentation, event, and test changes address the maintainer-bot keep_open feedback.
• fix: keep review-required kanban reblocks out of triage #61372 [closed] related — (+67/-4) — n/a: Closed reference implementation that keeps canonical review-required: reblocks in blocked; it remains relevant because it overlaps the salvageable reblock exemption in fix(kanban): enforce review change loop before completion #58695 and was intentionally moved to the author’s fork with no upstream merge intended.
• fix(kanban): persist pass-loop state on task rows #63723 related — (+1636/-39) — n/a: Adds broad persistent PASS-loop and delivery state rather than a configurable recurrence limit; its contributor keep_open review finds the completion_blocked_% predicate overbroad and the routing change unsupported without a current-main reproduction, so only a narrowly typed PASS-loop signal is salvageable.
• fix(kanban): fail closed invalid dependency waits #64585 related — (+124/-31) — n/a: Addresses the same parentless-dependency redispatch cause as fix: stop dependency-wait kanban promotion loops #61366 but fails closed in blocked; consistent with its keep_open verdict, it still needs CLI/tool-schema recovery guidance and interface coverage before this alternative can be retained.
• fix(kanban): guard dependency block after successor creation #66632 related — (+33/-0) — n/a: Rejects dependency blocks whenever a successor exists, but the contributor keep_open review shows this breaks valid P -> A -> C graphs and lets ValueError escape the CLI; salvage requires checking unsatisfied incoming parents and adding CLI regression coverage.
• fix(kanban): freeze repeated identical failures #68027 related — (+1114/-86) — n/a: Replaces kind-only counting with a normalized failure fingerprint, third-identical-outcome freeze, audited rearm, and guards across automatic paths; consistent with the maintainer-bot keep_open verdict finding no verified blocking defect, it is a broader failure-loop hardening lane and does not provide the requested configurable threshold.

Duplicates

#59536 and #59645 are later implementations of #59367. #61372 overlaps the review-reblock portion of #58695, while #61366 and #64585 address the same parentless-dependency cause with different terminal states.

Suggested consolidation

Author action: rebase #59367 onto current main, preserving its boolean-rejection tests and validated routing/event behavior; close #59536 and #59645 as duplicates of #59367 despite their keep_open reviews because their visible diffs retain the validation/default-registration defects summarized above, and close #66909 as a duplicate of #59333. Keep #68027 open with the salvage path of its durable fingerprint, audited rearm, and automatic-path guards; keep #59906, #61366/#64585, and #63723 in separately scoped salvage lanes addressing their recorded reviews, re-scope #60476 away from #59333, narrow #58695 to the safe review-reblock exemption unless an authorized ordered verdict source is added, and retain #61372 only as the closed reference.

Cross-PR triage: Reviewed 13 pull requests and 2 issues in this complex. Each diff was read against this issue; Assessment working set: 301 kB of PR diffs, 27 kB of issue/PR text, 17 kB of discussion (19 comments), 10 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.