Repository navigation
refactor: retire compliance_markers telemetry for claude sessions - #170
Merged
Merged
Conversation
The counter read 0 for RETRY-BUDGET, FETCH-PIVOT, DELEGATED and DIRECT-OK on 2026-10-29 across ~22 retry loops, while agents were following the rules: they made the pivot and skipped typing the line. An 80-minute loop and a clean session produced the same zero. omp retired the field on 2026-08-08; claude never followed. session-stats.sh stops computing it, SESSION_TRIAGE.md judges tool_loop severity on the transcript (same wording as the omp addendum), PROMPT.md ignores the field. Older findings keep it unread.
… and update the plan row Auditor seat (debate tight round 1, P2): a catch-up rebuild reuses findings JSONs triaged under the old rule, where a loop with no RETRY-BUDGET line was the serious finding. L2 now re-judges any tool_loop whose what cites a missing marker from its evidence quote. Swept prompts/ for other marker-severity text: none. omp-review round 1 (P3): docs/plans/2026-10-03-omp-adapter.md:145 said claude still computes the field.
drycode
added a commit
to drycode/cc-autodream
that referenced
this pull request
Nov 6, 2026
Bring the fork level with upstream d3ea856 (STRML#170-STRML#184) so the live install stops drifting: L2 counts from a per-session roll-up instead of a 200-entry sample, sandbox-shape denies are counted rather than scored as friction, changelog caches fetch anonymously and warn when every source fails, and OMP skill reads count as invocations. The tree is identical to upstream; the fork carries no content of its own. install.sh was re-run to link the new rollup.sh helper. Solves: fork 9 commits behind upstream Tests: tests/run-parallel.sh 3241 pass; same 7 fail on 599e3ca (TMPDIR //)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Retires
compliance_markersfor claude sessions. omp dropped it on 2026-08-08 and this side never followed (docs/plans/2026-10-03-omp-adapter.md: "ported for omp, not for claude").The counter read 0 for all four markers on 2026-10-29 across a day with ~22 retry loops. The agents were following the rule: a rush-ops session "read the failure-discipline leaf and the marker script instead of retrying" and switched gates, then never typed the
RETRY-BUDGET:line. So an 80-minute GUI-input loop and a clean session produced the same zero. Sam picked option B in the 2026-10-29 triage: drop the telemetry and judge retry discipline from the transcripts.bin/session-stats.shcompliance_markers(and the$assistant_textit alone used)prompts/SESSION_TRIAGE.mdtool_loopseverity judged on the transcript (same wording asadapters/omp/triage.md)prompts/PROMPT.mdtests/run-all.shOlder findings JSONs keep the field and both prompts say to leave it unread. The
failure-discipline.mdrule itself is unchanged.Test plan
tests/run-parallel.sh 4: 3097 passed, 4 failed. The four are inreview-cmux(host has cmux),adapter-claude(psunavailable in the sandbox) and theprune-worktreeswatchdog timing. None of those suites touch a changed file.session-stats.shemittedcompliance_markersbefore, not after./debate:run tight(repo rule for prompt andbin/changes)