manage-automations

Manage Automations

Review pending nominations and manage the automation lifecycle.

Process

Step 1: Load Nominations

Read all files in ~/.claude/projects/<project>/memory/nominations/. Parse YAML frontmatter to extract: id, confidence, type, target, pattern_type, auto_merged.

Step 2: Group and Present

Present nominations grouped by type, highest confidence first:

Auto-merged (for awareness): Show items that were auto-merged during mining. No action needed unless user wants to revert.

New skills/agents/hooks (require approval): For each, show:

• Confidence score and pattern type
• Evidence (message excerpts from source sessions)
• Proposed content
• Options: [accept] [edit] [reject] [defer]

New memories/rules (require approval for 50-79 confidence): Same format as above.

Updates to existing (show diff): For each, show:

• The existing file content
• The proposed changes (diff format)
• Options: [accept] [edit] [reject] [defer]

Step 3: Execute Actions

Accept: Write to the appropriate target location based on type:

Type	Target Location
feedback memory	~/.claude/projects/<project>/memory/knowledge/feedback_<topic>.md + update MEMORY.md
reference memory	~/.claude/projects/<project>/memory/knowledge/reference_<topic>.md + update MEMORY.md
rule	~/.agents/rules/<name>.md
claude_md	Append to relevant section in CLAUDE.md
skill	~/Projects/patryks-treadmill-claude-plugins/plugins/<best-fit-plugin>/skills/<name>/SKILL.md
agent	Same plugin structure, under agents/
hook	Plugin hooks config or ~/.claude/hooks/

After writing, delete the nomination file.

Edit: Present the proposed content for editing. User modifies, then write the modified version.

Reject: Delete the nomination file. No further action.

Defer: Leave the nomination file in place. It will appear again next time.

Step 4: Update Flow (for extensions)

When a nomination targets update:<existing-file-path>:

1. Read the existing file
2. Show the current content
3. Show the proposed updated content (generated by mining skill)
4. Present a diff between current and proposed
5. User approves → overwrite with updated content
6. Always show diff, even for memories (safety measure)

Step 5: Plugin Selection (for new skills/agents)

When creating a new skill or agent, determine the best-fit plugin:

1. Analyze the skill's domain (testing? workflow? quality? infrastructure?)
2. Match against existing plugins in treadmill:
   • kibana-testing-tools — testing, coverage, migration
   • kibana-code-quality-suite — quality, security, accessibility
   • kibana-dev-workflow-tools — git, PR, workflow
   • kibana-infrastructure-ops-tools — infra, monitoring, cross-repo
   • ci-babysitter — CI/CD
   • ai-conversation-intelligence — memory, patterns, conversation analysis
3. If no clear fit, place in ai-conversation-intelligence as default
4. Ask user to confirm placement before writing

Step 6: Marketplace Evaluation

After accepting any skill or agent nomination:

• Automatically invoke the marketplace-advisor skill to evaluate marketplace readiness
• If score >= 10, suggest publishing to treadmill marketplace
• If score 7-9, note improvements needed for future publishing
• If score < 7, skip silently

Step 7: Report

After all nominations are processed:

• Items accepted (with file paths)
• Items rejected
• Items deferred (still pending)
• Marketplace publishing actions taken or suggested
• Suggest re-running /mine-patterns if significant time has passed