refactor: workflow guidance fold-ins + dead-telemetry removal (net −107)#41
Merged
Merged
Conversation
Distill Anthropic's 'A harness for every task' write-up into the places that already decide when to orchestrate — no new files or skills. - orchestrate: failure-mode triggers (agentic laziness / self-preferential bias / goal drift), pattern vocabulary, quick workflows, token budgets, quarantine pattern; fix stale 'workflow' trigger keyword (now 'ultracode') - CLAUDE-FULL: failure-mode trigger in the delegation decision tree - oracle: pairwise/tournament judgment over absolute scores for close/taste calls - MANUAL: workflow token-budget directive - nuclear-review: example workflow reframed as a template - ship: fix duplicate 'Step 8' header Rejected the /ship -> bun run proof consolidation: proof is repo-local and ship runs in any project, so the cut would regress it elsewhere. Net +21 LOC; consolidation skill-merges deferred to a separate deliberate pass.
track-tldr.ts accumulated an *estimated* token-savings counter that only tldr-stats.ts read — to print a vanity box at session end, then delete the file. Closed read-loop, no external consumer; removing it loses only the end-of-session estimate display (TLDR itself is unaffected). Verified load-bearing and KEPT (not telemetry): swarm-log.ts feeds /review-batch (review-batch.ts reads ~/.claude/swarm.log); session-title.ts powers 'claude --resume <name>'. Removes both scripts, their hook registrations (PostToolUse mcp__tldr, SessionEnd tldr-stats), the integration test, and doc rows. handoff.ts create still runs at SessionEnd.
arzafran
force-pushed
the
refactor/workflow-guidance-consolidation
branch
from
65fb327 to
5f4c414
Compare
arzafran
added a commit
that referenced
this pull request
Jun 3, 2026
Versions work that had accumulated in [Unreleased] without a bump: team-knowledge repo migration (#38), nuclear-review maintainability pass, upstream-sync cron retirement + CC 2.1.161 sync (#40), dynamic-workflow guidance + dead-telemetry removal (#41), docs accuracy pass + USAGE.md removal (#42). VERSION 11.16.0 → 11.17.0; CHANGELOG [Unreleased] promoted to [11.17.0].
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
1 participant
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.
What this does
Folds Anthropic's dynamic-workflows learnings into the guidance that already decides when to orchestrate, and pays for the additions by deleting a dead telemetry loop — net −107 LOC, zero capability lost. Two commits, each independently valid.
1. Workflow nuggets (all fold-ins — no new files, no new skills)
skills/orchestrate/SKILL.md— the dynamic-workflows section now leads with the trigger (the three single-window failure modes: agentic laziness, self-preferential bias, goal drift) instead of task type; names the canonical shapes (classify-and-act, fan-out-and-synthesize, generate-and-filter, tournament, loop-until-done); adds quick workflows, token budgets, the quarantine pattern; fixes the staleworkflow→ultracodetrigger keyword (renamed in CC 2.1.160).CLAUDE-FULL.md— failure-mode trigger added to the delegation decision tree.skills/oracle/SKILL.md— compare mode prefers pairwise/tournament judgment over absolute weighted scores for close/taste calls.MANUAL.md— workflow token-budget directive;skills/nuclear-review/SKILL.md— example workflow reframed as a template to adapt.skills/ship/SKILL.md— fixed duplicateStep 8header.2. The cut (net-negative)
Removed the
track-tldr→tldr-statspair: a closed read-loop that accumulated an estimated token-savings counter only to print a vanity box at session end. No external consumer; TLDR itself is unaffected. Removes both scripts, their hook registrations, the integration test, and doc rows.Verified-and-spared — the real finding
Every other proposed cut failed verification and was kept. This is the answer to "are we bloated?": the skills are already lean (prior 37→26 passes), and almost everything is load-bearing.
/consolidate/nuclear-review/zero-tech-debt↔/refactorcontext: maininline vscontext: forkorchestrated)/oraclerisks ↔/verify/ship→bun run proofproofis repo-local;/shipruns in any projectswarm-log.ts/review-batch(review-batch.tsreads~/.claude/swarm.log)session-title.tsclaude --resume <name>The only honest LOC left to cut was dead telemetry.
Test Plan
bun run typecheckcleanbun test— 466 pass, 0 failbun run lint+bun run lint:skillscleanconfig/40-hooks.jsonvalid JSON; composedsettings.jsonhas 0 refs to deleted scripts