Skip to content

Monitoring & Debugging

See What's Running

bash
# State of any ADW run
cat agents/<adw-id>/adw_state.json

# Raw agent output for a phase
cat agents/<adw-id>/sdlc_planner/raw_output.jsonl | tail -1 | python3 -m json.tool

# All active ADW IDs
ls agents/

GitHub Comments Are Your Real-Time Log

The bot posts to the issue thread at every stage. Check the issue for:

  • Start confirmation with ADW ID
  • Phase completion messages
  • Review findings with screenshots (if Cloudflare R2 configured)
  • PR link

This is intentional — Mike's dashboard and GitHub Issues are the primary monitoring interface, not terminal output.

ADW State File

Every run writes agents/<adw-id>/adw_state.json with the current phase state:

json
{
  "adw_id": "a3f9k2m1",
  "issue_number": 42,
  "branch_name": "feat-42-a3f9k2m1-add-csv-export",
  "plan_file": "specs/issue-42-adw-a3f9k2m1-sdlc_planner-add-csv-export.md",
  "issue_class": "/feature",
  "phase": "build",
  "status": "running"
}

Use this to resume any failed phase:

bash
# Re-run build phase from existing plan
uv run adw_build.py 42 a3f9k2m1

# Re-run test phase
uv run adw_test.py 42 a3f9k2m1

# Re-run review phase
uv run adw_review.py 42 a3f9k2m1

Common Errors

"Agent execution failed"

bash
# Read what Claude actually did
cat agents/<adw-id>/sdlc_planner/raw_output.jsonl | python3 -m json.tool | tail -50

# Enable debug mode
export ADW_DEBUG=true
uv run adw_plan_build.py <issue>

"Missing plan file"

bash
# Re-run the plan phase
uv run adw_plan.py <issue>

"GitHub auth failed"

bash
gh auth status
gh auth login

"Claude Code CLI is not installed"

bash
which claude  # Check if installed
claude --version
# If missing, reinstall via npm
npm install -g @anthropic-ai/claude-code

"Missing GITHUB_PAT"

bash
# Use gh CLI auth token as fallback
export GITHUB_PAT=$(gh auth token)

Debug Mode

bash
export ADW_DEBUG=true
uv run adw_plan_build.py 123  # Verbose output

Debug mode logs every API call, every git operation, and every Claude Code session event.

Output File Structure

Each ADW agent writes to its own directory:

FileContents
cc_raw_output.jsonlRaw streaming output (one JSON per line)
cc_raw_output.jsonAll messages as a parsed JSON array
cc_final_object.jsonLast entry — the final result object
custom_summary_output.jsonHigh-level summary: ADW ID, model, success, session ID

Checking Environment Variables

bash
# Verify required variables are set
env | grep -E "(GITHUB|ANTHROPIC|CLAUDE)"

# Verify GitHub auth
gh auth status

# Test Claude Code
claude --version

Cleanup

bash
# Remove a specific run workspace
rm -rf agents/<adw-id>

# Remove all run workspaces (keeps specs/)
rm -rf agents/*/

Agentics SOP — AI Developer Workflow & Agentic Coding Reference