sdlc-health

name	sdlc-health
description	Diagnostic health check for the AgentFlow pipeline. Verifies prerequisites, checks connectivity, reports pipeline status. In plugin mode, also checks plugin installation, agents, hooks, and inter-agent communication.

/sdlc-health

Run a comprehensive health check on the AgentFlow pipeline. Report status for every subsystem and provide actionable fix instructions for any issues found.

Execution

Run all checks below in order. Collect results, then print the summary at the end.

Checks

1. Prerequisites

Verify each of these exists and is functional:

• Claude Code CLI -- run claude --version to confirm it is accessible
• git -- run git --version
• gh (GitHub CLI) -- run gh --version, then gh auth status to confirm authentication
• Node.js -- run node --version
• npm -- run npm --version
• Conventions file -- check ~/.claude/sdlc/conventions.md exists
• Skill files -- check ~/.claude/skills/ contains: spec-to-asana.md, sdlc-worker.md, sdlc-orchestrate.md, sdlc-stop.md, sdlc-health.md, sdlc-demo.md
• Prompt files -- check ~/.claude/sdlc/prompts/ contains: decompose.md, research.md, build.md, review.md, test.md

For any missing file, report exactly which file is missing and suggest running ./setup.sh from the agentflow repo.

2. PM Tool Connectivity

Test the Asana MCP connection:

• MCP responding -- call the Asana MCP get_me tool. If it responds with user info, the connection is live.
• Can list projects -- call get_projects and confirm at least one project is returned
• Can read tasks -- pick any project, call get_tasks for it
• Can post comments -- do NOT actually post a comment; just confirm the add_comment tool is available in the MCP toolset

If MCP is not responding, report: "Asana MCP server not connected. Check your MCP configuration in Claude Code settings."

3. Pipeline Status

Search for projects with [SDLC] in the name. For each one found:

• Count tasks per stage: Backlog, Research, Build, Review, Test, Integrate, Done, Needs Human
• List active workers: which slots (T2-T5) have assigned tasks, and which tasks
• Find the Status task (pinned task with [LAST_SWEEP:] in description)
  • Extract the last sweep timestamp
  • Calculate time since last sweep
  • Classify: HEALTHY (<20 min), WARNING (20-60 min), CRITICAL (>60 min)
• List any tasks in "Needs Human" (task name and code)
• Report total estimated cost across all active tasks (sum [COST:~$N] values)
• Check for active merge locks ([MERGE_LOCK:] in Status task)
• Check for active sweep locks ([SWEEP:RUNNING] in Status task)

If no [SDLC] projects found, report: "No AgentFlow projects found. Create one with: /spec-to-asana"

4. Crontab Status

• Crontab entry -- run crontab -l and check for a line containing agentflow-cron.sh
• Cron wrapper exists -- check ~/.claude/sdlc/agentflow-cron.sh exists and is executable
• Log file exists -- check /tmp/agentflow-orchestrate.log exists
• Recent log entry -- if log exists, read the last 5 lines. Check if the most recent entry is less than 30 minutes old.
• Last entry successful -- check that the last log entry does NOT contain "ERROR"

If crontab is not configured: "Crontab not set up. Run: ./setup.sh --with-cron"

5. Git Status

Run these in the current working directory:

• Not on main -- git branch --show-current should NOT be main or master (workers should be on feature branches)
• Clean working tree -- git status --porcelain should be empty (uncommitted changes block worktree creation)
• Remote configured -- git remote get-url origin should return a valid URL
• Remote accessible -- git ls-remote --exit-code origin HEAD should succeed

6. LEARNINGS.md Status

If a LEARNINGS.md file exists in the current project:

• Count the number of lines
• If approaching 50 lines: report as INFO
• If over 50 lines: report as WARNING ("LEARNINGS.md is over the 50-line cap. Consider archiving old entries.")

7. Plugin Status (Plugin Mode Only)

If running as a Claude Code plugin:

• Plugin installed — check if agentflow plugin is recognized
• Agent definitions — verify all 4 agents are loadable
• Hook definitions — verify all 3 hooks are loadable
• Team status — if sprint-team exists, report member health
• SendMessage working — verify inter-agent communication

Output:

Plugin:         [OK] v2.0.0, 4 agents, 3 hooks (or [N/A] Not in plugin mode)

Output Format

Print a clean summary:

AgentFlow Health Check
======================
Prerequisites:  [OK] All 12 checks passed (or [FAIL] Missing: <list>)
PM Tool:        [OK] Connected as <username> (or [FAIL] <error>)
Pipeline:       N project(s), M active tasks, K in Needs Human
Orchestrator:   [OK] Last sweep 5 min ago (or [FAIL] No sweep in 60 min!)
Crontab:        [OK] Configured, last run 12 min ago (or [FAIL] Not set up)
Git:            [OK] Clean, on branch feature/xyz (or [WARN] Uncommitted changes)
LEARNINGS.md:   [OK] 23 lines (or [WARN] 52 lines -- over 50-line cap)
Plugin:         [OK] v2.0.0, 4 agents, 3 hooks (or [N/A] Not in plugin mode)

Then list any issues, sorted by severity:

Issues Found:
  [CRITICAL] No orchestrator sweep in 65 minutes -- check crontab and /tmp/agentflow-orchestrate.log
  [CRITICAL] Asana MCP not responding -- check MCP server configuration
  [WARNING]  2 tasks in Needs Human: APP-003, APP-007
  [WARNING]  GitHub CLI not authenticated -- run: gh auth login
  [INFO]     LEARNINGS.md is 48 lines (approaching 50-line cap)
  [INFO]     No AgentFlow projects found -- create one with /spec-to-asana

If zero issues found, print:

No issues found. Pipeline is healthy.

Fix Instructions

For every issue reported, include a one-line fix command or instruction:

Issue	Fix
Claude CLI not found	Install from https://claude.ai/code
git not found	Install git for your platform
gh not installed	Install from https://cli.github.com/
gh not authenticated	Run: gh auth login
Missing skill/prompt files	Run: cd /path/to/agentflow && ./setup.sh
Asana MCP not responding	Check MCP server config in Claude Code settings
No SDLC projects	Run: /spec-to-asana with a SPEC.md in your project
Crontab not configured	Run: ./setup.sh --with-cron
No recent sweep	Check: tail -20 /tmp/agentflow-orchestrate.log
Sweep errors in log	Check: grep ERROR /tmp/agentflow-orchestrate.log | tail -5
On main branch	Create a feature branch: git checkout -b feature/your-feature
Uncommitted changes	Commit or stash: git stash
LEARNINGS.md over cap	Archive old entries to LEARNINGS-archive.md