Merged
Conversation
Introduces a polish step (Step 6) that runs after all tasks complete, performing multi-pass analysis to catch cross-task issues (DRY violations, naming drift, integration gaps) and orchestrating fixes via the existing executor and reviewer agents. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Replace brittle git diff command with natural language intent for identifying implementation scope - Flesh out fixed analysis passes — each sub-agent now receives file list and detailed analysis criteria, with structured findings expected - Move test rules into fix cycle as pass-through to executor/reviewer rather than standalone section (polish agent doesn't write code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
… subagent_type The claude-technical-workflows: prefix is a marketplace namespace only available when the package is installed. Match existing convention from invoke-executor.md and invoke-reviewer.md: reference agents by name and .claude/agents/ path. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The polish agent needs to know HOW to read tasks from the plan — plans may be stored in external tools (e.g. Linear) where simply reading the file is insufficient. Pass the format's reading.md adapter so the agent can retrieve the full task landscape regardless of storage format. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Replace numbered list with lettered stages (A-D) matching task-loop conventions. Add STOP gates after executor and reviewer invocations, explicit routing on STATUS/VERDICT, and clear sequential flow. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Replace explanatory prose + separate sections with sequential Steps (1-4) and lettered stages (A-F) within the discovery-fix loop. STOP gates after every agent dispatch, explicit routing on STATUS/VERDICT, inline analysis criteria at point of use. Matches task-loop and SKILL.md conventions — deterministic orchestration with creativity scoped to analysis sub-agents only. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Plans implement specific features within a larger system. Polish must stay within the plan's boundary — don't flag missing features that belong to other plans (e.g., "lacks authentication" when auth is a separate plan). Look inward at what this plan built, not outward at what the broader system might be missing. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add stage A (Cycle Gate) as the loop entry point with inline cycle count logic — increment, check exit conditions, route to B or Step 4 - Fix stage C (Dynamic Analysis) — check if needed BEFORE dispatching, not after the STOP gate - Renumber stages A-G with G always returning to A (proper loop) - All routing now flows correctly: gate → dispatch → synthesize → fix → complete → gate Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Syncs flowchart data with recent changes: - Add implementation-polish agent (flowchart, sidebar, desc, source map) - Update skill-implementation: - Add fix_gate_mode routing in Review Changes - Add Step 6 Polish (checkpoint → polish agent → approval gate) - Task loop now exits to polish, not directly to completion - Update implementation-task-reviewer: - 6 dimensions (added codebase cohesion) - Fix recommendations (FIX/ALTERNATIVE/CONFIDENCE) - Add read-context step in flowchart - Update descriptions for skill-implementation and reviewer Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
leeovery
added a commit
that referenced
this pull request
Feb 6, 2026
…itions) V3's integration context created a "convention gravity well" where early design choices got documented as established patterns and later tasks faithfully propagated them. V2 without this mechanism won 21/23 tasks. Removes: integration context file accumulation, plan file access for executor, expanded exploration instructions, codebase cohesion review dimension, Convention Consistency quality section. Keeps: fix recommendations (PR #78), fix_gate_mode (PR #78), full re-attempt context (PR #77), polish agent (PR #80). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
6 tasks
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.
Summary
agents/implementation-polish.md) that performs holistic quality analysis after all tasks complete — catching cross-task issues like DRY violations, naming drift, integration gaps, and missing integration testsy/skip/comment), and safe discard via working tree resetTest plan
agents/implementation-polish.mdfollows existing agent conventions (frontmatter, tools, model, hard rules, structured output)invoke-polish.mdfollows the same pattern asinvoke-executor.mdandinvoke-reviewer.mdy/skip/comment routingskippath safely discards only polish changes via checkpoint resetcommentre-invocation path does not re-checkpoint🤖 Generated with Claude Code