Help us improve
Share bugs, ideas, or general feedback.
Explains, traces, audits, or explores code in read-only mode without modifying it. Answers questions like 'how does X work' or 'where is Y called' with evidence.
npx claudepluginhub bravekingzhang/agent-coding-playbook --plugin agent-coding-playbookHow this skill is triggered — by the user, by Claude, or both
Slash command
/agent-coding-playbook:acp-investigateThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Use this skill when the user wants to understand code, not change it.
Investigates code by tracing execution paths, mapping data flow, and diagnosing root causes. Outputs an analysis report with architecture overview and findings.
Traces actual code paths in codebases to analyze architecture, data flows, integrations, patterns, and provide recommendations across 5 evidence-based iterations. Use for deep 'how does X work' queries or complex system reviews.
Deep-dive into a codebase question, failure, or architectural concern without making changes
Share bugs, ideas, or general feedback.
Use this skill when the user wants to understand code, not change it.
The goal is not to fix a problem.
The goal is to produce an accurate, evidence-based explanation of what the code does, where it lives, and why.
If the user is asking a question rather than asking for a change, stay in this skill until they explicitly ask for a code change.
While in this skill:
If you discover a bug or unrelated issue during investigation, surface it in the report. Do not silently patch it.
Before searching, write down:
A vague question produces a vague answer. Ask for clarification if the question is ambiguous.
For anything beyond a single grep, prefer the Explore subagent over loading files into the main session.
Explore is faster, cheaper, and keeps the main context clean. Reserve direct file reads for the small set of files you actually need to quote.
Good queries to delegate:
renderAvatar and group by file."/api/orders and trace the next 2 layers of calls."process.env.STRIPE_KEY."For the small set of files you do read directly:
Every claim in the report should be backed by:
Do not paraphrase code into a confident summary that the code does not actually support.
Produce a structured answer. Keep it as short as the question allows.
Question:
- ...
Answer:
- ...
Evidence:
- file:line — what it shows
- file:line — what it shows
Open questions:
- ...
Things noticed (not fixed):
- ...
If the user, after seeing the investigation, asks for a fix or change, exit this skill and switch to the appropriate one (acp-bug-fix, acp-refactor, acp-feature-add).
Do not start editing while still in investigate mode.