Claude Code Plugins

Community-maintained marketplace

Feedback

agent-investigator

@shikihane/polydev
1
0

SUB-AGENT ONLY: Use when spawned for read-only research - generates report file and outputs [AGENT_DONE] marker

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name: agent-investigator description: SUB-AGENT ONLY: This skill should be used when spawned for read-only research - generates report file and outputs [AGENT_DONE] marker. version: 0.1.0

Agent Investigator

Execute investigation tasks and generate structured reports.

Note: This skill is used by sub-agents, automatically activated when started by spawn-agent.sh.

Important: Sub-Agent Context

This is a sub-agent started by the main agent via spawn-agent.sh.

Do NOT call any polydev scripts. The responsibilities are:

  1. Understand the investigation task (from startup prompt)
  2. Execute investigation (search, read, analyze code/docs)
  3. Generate report (write to specified file)
  4. Output completion marker (let main agent know task is done)

The main agent detects [AGENT_DONE] output via terminal capture.

Completion Marker Format (Must Follow)

When task is complete, output this exact format to terminal:

[AGENT_DONE]
report: <full path to report file>
timestamp: <ISO timestamp, e.g., 2025-01-05T10:30:00Z>
summary: <summary in 20 words or less>

Example:

[AGENT_DONE]
report: ./.agent-reports/auth-analysis.md
timestamp: 2025-01-05T10:30:00Z
summary: JWT auth, 3 middlewares, 2 security issues found

The main agent detects your completion via grep "\[AGENT_DONE\]".

Report Template

# Investigation Report: <Topic>

Generated: <ISO timestamp>

## Summary

<3-5 sentences summarizing key findings - this is what the main agent sees first>

## Findings

### 1. <Finding Title>

<Detailed explanation>

**Key Code:**
```<language>
// path/to/file.ts:123
<relevant code snippet>

2.

Key Files

File Lines Description
src/auth/jwt.ts 45-67 JWT validation logic
src/middleware/auth.ts 12-34 Auth middleware

Recommendations

  1. -
  2. -

Appendix (Optional)

<Additional technical details, code snippets, references>


---

## Execution Flow
  1. Read task description |
  2. Plan investigation steps |
  3. Execute investigation (search, read, analyze)
    • Use Glob to find files
    • Use Grep to search patterns
    • Use Read to read code |
  4. Organize findings, write report |
  5. Write report to specified file |
  6. Output [AGENT_DONE] marker |
  7. Done (main agent reads report)

---

## Prohibited Actions

DO NOT wait for user input (running in background) DO NOT output lots of process info to terminal (wastes main agent's tokens) DO NOT forget to output [AGENT_DONE] marker DO NOT modify code (unless task explicitly requires) DO NOT call any polydev scripts (sub-agent context)

DO keep process info in head or write to log DO write final results to report file DO output only key progress and completion marker to terminal DO structure report for quick scanning


---

## Terminal Output Example

Ideal terminal output should be very concise:

Starting investigation: Authentication mechanism analysis

Searching auth-related files... Found 12 related files

Analyzing key code...

  • src/auth/jwt.ts
  • src/middleware/auth.ts
  • src/routes/login.ts

Writing report...

Report generated: ./.agent-reports/auth-analysis.md

[AGENT_DONE] report: ./.agent-reports/auth-analysis.md timestamp: 2025-01-05T10:30:00Z summary: JWT auth, 3 middlewares, 2 security issues found


---

## Communication with Main Agent

There is no direct communication channel between sub-agent and main agent. The only ways to communicate are:

1. **Report file** - Main agent reads the generated report
2. **[AGENT_DONE] marker** - Main agent detects completion via terminal output
3. **summary field** - Main agent quickly decides if detailed reading is needed

Keep reports concise and structured so main agent can quickly extract key information.

---

## Rule Reflection (Check Before Completion)

**Trigger conditions** (all must be met):
1. Encountered **environment/compatibility/parameter usage** issue
2. Issue **will recur when new Agent executes**
3. Issue **has been solved** with a clear solution

**Action:** Write file to `.agent-memory/proposed-rules/<issue-summary>.md`

**Format:**
```markdown
# <Issue Summary>

## Problem
<Describe the issue and trigger conditions>

## Solution
<Specific solution>

## Example
\`\`\`bash
# Wrong approach
...

# Correct approach
...
\`\`\`

Do NOT trigger for:

  • Business logic issues
  • One-time issues
  • Uncertain if generalizable