From T.L-AutoDevelop
Analyze a completed or in-progress AutoDevelop session and give honest, constructive feedback on friction points, failure patterns, inefficiencies, and improvement opportunities in the pipeline.
How this skill is triggered — by the user, by Claude, or both
Slash command
/T.L-AutoDevelop:pipe-feedbackThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Your job is to observe, analyze, and give honest feedback on the AutoDevelop pipeline based on what actually happened in this session. You are not here to summarize progress — you are here to find problems, rough edges, and improvement opportunities that the developer (who built this pipeline) would want to know about.
Your job is to observe, analyze, and give honest feedback on the AutoDevelop pipeline based on what actually happened in this session. You are not here to summarize progress — you are here to find problems, rough edges, and improvement opportunities that the developer (who built this pipeline) would want to know about.
Be specific. Be direct. Praise what worked well too — balanced feedback is more useful than a list of complaints.
Allowed tools: Read, Glob, Grep, Bash.
Do not modify any files. Do not run the scheduler. Do not start tasks.
Find the scheduler state and recent run artifacts:
<repo-root>/.claude-develop-logs/scheduler/
<repo-root>/.claude-develop-logs/runs/
If no run artifacts exist, report that there is no session data to analyze and stop.
Read:
state.json or equivalent)For each recent run, read if present:
timeline.json — phase transitions and milestonesscheduler-snapshot.json — task state at the time of the runFrom the artifacts, build a factual picture of the session:
merged or completed_no_change)?retry_scheduled, manual_debug_needed, or completed_failed_terminal?Look specifically for:
Retry patterns
FIX_PLAN_INSUFFICIENT appear on tasks that seem straightforward? That may indicate an overly strict plan evaluator or missing codebase context.Stall patterns
Orchestration friction
C:\Temp\) to parse scheduler output? That indicates the scheduler output is too large or lacks a compact summary mode.Context quality
Be equally specific about things that functioned smoothly:
Structure the report as follows:
One short paragraph: what ran, what succeeded, what failed, overall health.
Bullet list. Specific and honest — skip this section if nothing stood out positively.
For each issue:
Prioritize by impact. Do not pad with minor nitpicks unless the serious issues list is short.
Things that didn't break this session but could break at scale or with more tasks. Flag them clearly as early warnings, not confirmed bugs.
One or two sentences. Honest. If the session was rough, say so. If it went well despite a few bumps, say that too.
npx claudepluginhub tleiott/t.l-marketplace4cctoolz --plugin T.L-AutoDevelopPerforms post-pipeline retrospectives: parses logs, counts productive vs wasted iterations, identifies failure patterns, scores runs, suggests fixes to skills/scripts.
Root cause analyzer for AI coding sessions. Detects scope drift, rework patterns, and root causes to improve future prompts and repo health.
Audits Claude Code session logs for PBR workflow compliance (STATE.md updates, hooks, commits, skills) and UX quality (flows, friction, expectations). Supports date ranges and modes.