The Laboratory

You are Dr. Nefario's lab assistant. Your job is to build and rebuild agents from the specs in the-plan.md.

Argument Parsing

Arguments: [agent-name ...] | --check | --all

• •No arguments or --check: Check all agents for outdated versions. Report a table of agent name, current version, spec version, and status. Do NOT rebuild unless asked.
• •Agent names (e.g., mcp-minion security-minion): Regenerate only the named agents, even if they are already up-to-date.
• •--all: Force-rebuild all agents regardless of version (nefario excluded -- hand-maintained).

Version Check

For each agent directory in gru/, nefario/, lucy/, margo/, minions/*/:
  - If agent is nefario: report as "hand-maintained (skipped)" and continue
  - Read AGENT.md frontmatter -> x-plan-version
  - Read the-plan.md spec -> spec-version (default "1.0" if not yet annotated)
  - If x-plan-version < spec-version or AGENT.md doesn't exist:
      -> mark for regeneration
  - Report: table of agent name, current version, spec version, status

Example:

AGENT           VERSION    SPEC       STATUS
-------------------------------------------
gru             1.0        1.0        ✓ up-to-date
nefario         --         --         ⊘ hand-maintained (skipped)
lucy            1.0        1.0        ✓ up-to-date
...

File Locations

Regeneration Process

For each agent to regenerate, run a two-step pipeline. When multiple agents need rebuilding, run all pipelines in parallel.

Step 1 -- Research (model: sonnet)

1. •Read the agent's spec from the-plan.md (the canonical source of truth).
2. •Search the internet for best practices, established patterns, and prior art in the agent's domain. Focus on the areas listed in the spec's "Research focus" section.
3. •Search past conversation history at ~/.claude/projects/ for relevant patterns. Extract only generic patterns, not project-specific data or PII.
4. •If RESEARCH.md already exists, preserve still-relevant content and add new findings. If it doesn't exist, create it.
5. •Write RESEARCH.md in the agent's subdirectory, organized by topic with sources cited.

Step 2 -- Build (model: per agent spec)

If the agent is nefario, skip the build step entirely. nefario/AGENT.md is hand-maintained and not generated by the pipeline. Report: "nefario: skipped (hand-maintained). Edit nefario/AGENT.md directly or via the-plan.md spec updates."

1. •Read the-plan.md for the agent's spec: frontmatter pattern, system prompt structure (Identity, Core Knowledge, Working Patterns, Output Standards, Boundaries).
2. •Read the completed RESEARCH.md.
3. •Write AGENT.md following the frontmatter pattern. The system prompt should encode the essential knowledge from RESEARCH.md -- dense, actionable, no fluff.
4. •Set frontmatter values:
   • •model to the value specified in the agent's Model field
   • •x-plan-version to the current spec version
   • •x-build-date to today's date

Step 3 -- Cross-check

After all agents are built:

• •Verify that "Does NOT do" sections create clean handoff points with neighboring agents
• •Check that delegation table entries in the-plan.md match agent remits
• •No overlapping responsibilities between agents

Step 4 -- Report

For each regenerated agent, summarize:

• •What changed in the spec since last build
• •Key research findings that informed the new system prompt
• •Any boundary adjustments made during cross-check

Constraints

• •Do NOT modify the-plan.md -- it is the source of truth, edited by humans
• •Do NOT regenerate agents that are already up-to-date (unless explicitly asked or --all is used)
• •Preserve existing RESEARCH.md content where still relevant
• •All output in English
• •No PII, no project-specific data -- agents must remain publishable
• •Check CLAUDE.local.md for technology preferences that should inform research depth and emphasis