Troubleshooting

This page covers the most frequent issues OrchestKit users encounter, ordered by how often they come up. For hook-specific debugging, see Hook Debugging. For broader questions about how OrchestKit works, see the FAQ.

1. Plugin Not Found After Installation

Symptom: Claude Code does not recognize /ork: commands or shows "no matching skill."

Cause: The plugin is not installed in the right location, or Claude Code has not reloaded.

Fix:

# Verify the plugin is installed
claude plugins list

# If missing, reinstall
claude plugins install orchestkit

# If installed but not detected, restart Claude Code
claude --restart

2. Hooks Not Firing

Symptom: Expected behavior does not happen -- skills are not auto-suggested, dangerous commands are not blocked, or memory is not injected.

Cause: Hooks are disabled, hooks.json is missing, or the TypeScript bundles are not compiled.

Fix:

# Run diagnostics
/ork:doctor

# Check if hooks.json exists
ls src/hooks/hooks.json

# Rebuild hooks from TypeScript
cd src/hooks && npm run build

# Verify the compiled bundles exist
ls src/hooks/dist/

If /ork:doctor reports hooks as unhealthy, the most common cause is missing compiled bundles in dist/. The TypeScript source in src/hooks/src/ must be compiled before hooks can execute.

See Hook Debugging for deeper investigation steps.

3. Skills Not Loading

Symptom: Invoking /ork:skillname returns "skill not found" or agents do not have expected skills injected.

Cause: The skill is not listed in the plugin manifest, or the SKILL.md frontmatter is malformed.

Fix:

# Verify skill exists in source
ls src/skills/your-skill/SKILL.md

# Check the manifest includes the skill
grep "your-skill" manifests/ork.json

# Validate skill structure
npm run test:skills

# Rebuild the plugin
npm run build

Common frontmatter issues:

• Missing required name or description fields
• user-invocable: true missing for command skills
• Invalid YAML syntax (tabs instead of spaces, unclosed quotes)

4. Memory Not Persisting Across Sessions

Symptom: Decisions stored with /ork:remember do not appear in the next session.

Cause: The MCP memory server is not configured, or local memory files are not being written.

Fix:

# Check memory system health
/ork:memory status

# Verify local memory files exist
ls .claude/memory/

# Check if MCP memory server is running
claude mcp list

The memory system has 3 tiers. If Tier 1 (Graph) is not configured, decisions fall back to Tier 2 (Local files). If those are not writable, nothing persists.

5. Agent Spawn Failures

Symptom: Running /ork:implement or /ork:review-pr fails with "agent not found" or agents produce empty results.

Cause: Claude Code version is too old, the requested model is unavailable, or the agent definition file is malformed.

Fix:

# Check Claude Code version (requires >= 2.1.34)
claude --version

# Verify agent definition exists
ls src/agents/backend-system-architect.md

# Validate agent frontmatter
npm run test:agents

# Rebuild plugin
npm run build

If the Claude Code version is below 2.1.34, upgrade:

npm install -g @anthropic-ai/claude-code@latest

6. Build Errors After Editing src/

Symptom: npm run build fails after modifying skills, agents, or hooks.

Cause: A manifest references a skill or agent that does not exist, or a new file has invalid frontmatter.

Fix:

# Run the full test suite to identify the issue
npm test

# Common: skill referenced in manifest but not in src/
diff <(jq -r '.skills[]' manifests/ork.json | sort) \
     <(ls src/skills/ | sort)

# Common: hook referenced in hooks.json but not compiled
cd src/hooks && npm run build

7. Pre-Push Hook Hanging

Symptom: git push hangs on "Running quick lint tests..." and never completes.

Cause: The pre-push hook runs ./tests/run-all-tests.sh --lint, which can hang on large test suites or when external services are unreachable.

Fix:

For safe changes (docs, config, test fixes), bypass the hook:

git push --no-verify

For the permanent fix, investigate which test is hanging:

# Run tests with verbose output to find the stall
./tests/run-all-tests.sh --lint --verbose

8. Count Drift After Adding Skills or Hooks

Symptom: Tests fail with "expected 67 skills, found 69" or similar count mismatches.

Cause: Adding a skill or hook updates the source files but not the count references scattered across the codebase.

Fix:

# Run the count validation script
./tests/validate-counts.sh

# Update counts in all locations
# The script will tell you which files need updating

Files that reference counts and must stay in sync:

• CLAUDE.md (updated automatically by linter hooks)
• README.md
• manifests/ork.json
• tests/skills/structure/test-skill-md.sh
• tests/hooks/split-bundles.test.ts

9. Git Operations Blocked

Symptom: Git commands fail with "operation blocked by file-guard" or "branch protection prevents this action."

Cause: OrchestKit's safety hooks are preventing a potentially destructive operation.

Fix:

The dangerous-command-blocker hook blocks commands like git push --force, git reset --hard, and rm -rf by default. This is intentional.

If the operation is legitimate:

1. Check the hook output for which rule triggered
2. If it is a branch protection rule: create a feature branch instead of committing to main
3. If it is a file-guard rule: verify you are not modifying protected files (.claude/coordination/, secrets)
4. If the block is incorrect: the hook returns {"continue": false} with a reason -- share this in a GitHub issue

10. Slow Performance

Symptom: Skills take a long time to execute, sessions feel sluggish, or agent spawns are delayed.

Cause: Too many hooks executing synchronously, large context windows, or unoptimized hook bundles.

Fix:

# Check hook execution time
/ork:doctor

# Disable non-essential hooks temporarily
# Edit src/hooks/hooks.json and set "enabled": false on specific hooks

# Verify split bundles are working (should be 12 bundles, not 1 monolith)
ls src/hooks/dist/

Performance tips:

• OrchestKit uses 11 split bundles instead of a monolith to reduce load time
• 6 hooks use fire-and-forget async execution for non-blocking background work
• If a specific hook is slow, disable it temporarily and report the issue
• Context window usage above 85% triggers automatic result condensing in memory searches

Still Stuck?

• Run /ork:doctor for a full health diagnostic
• Check Hook Debugging for hook-specific issues
• Read the FAQ for common questions
• Open a GitHub issue with the error message and /ork:doctor output