Replace re-anchoring prose with early-completion delegation pattern (Fixes #121)#122
Replace re-anchoring prose with early-completion delegation pattern (Fixes #121)#122LeanAndMean wants to merge 4 commits intomainfrom
Conversation
…-fix (Part of #121) Mark Step 3 complete before the Skill invocation so the task list state (Step 3 complete, Step 4 pending) serves as the re-anchoring signal instead of post-delegation prose. Rename the step to "Delegate review fixes to feature-dev" to make the early-completion semantic self-evident, and switch phase naming from "Phase N:" to "Step 3.N:" for hierarchical parent-child visibility. Co-Authored-By: Claude <noreply@anthropic.com>
…(Part of #121) Co-Authored-By: Claude <noreply@anthropic.com>
PR Review: Replace re-anchoring prose with early-completion delegation patternCritical Issues (2 found)F1: CLAUDE.md "Delegation step completion" and "Skill and subagent delegations" overgeneralize to all Skill delegations The new "Delegation step completion" bullet states a blanket rule: "When a tracked step delegates to a Skill (via the Skill tool), mark the step complete before the Skill invocation." The "Skill and subagent delegations" bullet states: "For Skill delegations, the Skill creates hierarchical sub-tasks ( The old "Post-delegation re-anchoring" bullet was more carefully scoped with the parenthetical "Long-running Skill calls (especially Files: F2: CLAUDE.md states "no prose needed" but pr-review-fix.md adds explanatory prose The "Delegation step completion" bullet in CLAUDE.md explicitly states "the task list state is the re-anchoring signal, no prose needed." However, Files: Important Issues (3 found)F3: Step 3 marks complete before any preparation happens In Files: F4: Significant duplication of The Files: F5: "At least one sub-task must be created" requirement has no enforcement in the command file The CLAUDE.md "Delegation step completion" bullet states "At least one sub-task must be created" but Files: Suggestions (4 found)S1: Remove the explanatory sentences from S2: Simplify the "State transitions" bullet by replacing the multi-sentence delegation exception with a brief cross-reference to "Delegation step completion" (e.g., "Exception: delegation steps follow the early-completion pattern described in Delegation step completion below."). This keeps the bullet focused on the general rule. S3: Consider merging "Delegation step completion" and "Delegation phase tracking" into a single "Delegation conventions" bullet to reduce duplication of the S4: Move the forward-looking note about Strengths
Automated review by Claude Opus 4.6 using pr-review-toolkit (code-reviewer, comment-analyzer, code-simplifier) and mach10 (feature-completeness-checker) |
Independent AssessmentAssessing findings from the review comment. Finding ClassificationsF1: CLAUDE.md overgeneralizes early-completion to all Skill delegationsClassification: Genuine issue Verified. CLAUDE.md line 109 says "When a tracked step delegates to a Skill (via the Skill tool), mark the step complete before the Skill invocation" -- a blanket rule. Line 108 says "For Skill delegations, the Skill creates hierarchical sub-tasks" -- also blanket. But F2: CLAUDE.md states "no prose needed" but pr-review-fix.md adds explanatory proseClassification: Genuine issue Verified. CLAUDE.md line 109 states "the task list state is the re-anchoring signal, no prose needed." Yet F3: Step 3 marks complete before any preparation happensClassification: False positive This is the intended design, not a bug. The CLAUDE.md convention explicitly prescribes this pattern: early completion for delegation steps. The preparation work (gathering context, selecting findings) happens in Steps 1 and 2. Step 3 is named "Delegate review fixes to feature-dev" -- its purpose is the handoff, not execution. The Skill invocation context is assembled from data already collected. F4: Significant duplication of
|
| Classification | Count | Findings |
|---|---|---|
| Genuine | 2 | F1, F2 |
| False positive | 2 | F3, F5 |
| Nitpick | 1 | F4 |
| Deferred | 1 | S4 |
Independent assessment by Claude Opus 4.6
…iew findings (Part of #121) Co-Authored-By: Claude <noreply@anthropic.com>
Progress updateCommit 33f7854 addresses review findings F1 and F2, and applies the early-completion delegation pattern to
|
#121) The early-completion pattern removed the post-Skill "Mark Step N complete" line that previously served as a continuation signal. After context-heavy Skill delegations, the model stalled instead of resuming the outer command. Add a one-line continuation cue after each Skill invocation block and update the CLAUDE.md convention to require it. Co-Authored-By: Claude <noreply@anthropic.com>
Progress UpdateCommit: 8e33b2e — Add post-delegation continuation cue to prevent workflow stall What happenedDuring the What changed
This is a minimal one-line cue — not re-anchoring prose. It doesn't re-establish context about what happened; it simply tells the model what to do next. |
Summary
pr-review-fix.mdwith an early-completion pattern that marks Step 3 complete before the Skill invocation, using the task list state as the re-anchoring signal"Phase N:"to"Step N.M:"for hierarchical parent-child visibility in the flat task listTest plan
/mach10:pr-review-fixon a PR with at least one review finding and verify all 5 steps (0-4) show completed/mach10:pushnext-step suggestion is output after feature-dev delegation completes"Step 3.N:"naming in the task listFixes #121
Generated with Claude Code