Matilha Scout

When this fires

User wants to explore a feature's problem space before writing a spec. project-status.md shows current_phase: 0 or 10. The skill dispatches research subagents (market, user needs, competitive, technical, regulatory if relevant) in parallel and synthesizes their output.

Preconditions

• Feature slug provided (or inferable from prompt).
• matilha-scout lazy-bootstraps project structure; project-status.md is created if absent.

Execution Workflow

1. Lazy bootstrap: Ensure docs/matilha/research/ exists (mkdir -p). If project-status.md missing, write a minimal stub (current_phase: 0, phase_status: in_progress, project_slug: <from cwd or prompt>, next_action: "matilha-scout running"). Read project-status.md; verify current_phase ≤ 10 (if present and higher, warn user — scout is Phase 10 territory).
2. Dispatch 3-5 research subagents via Task tool IN PARALLEL (one Task call per scope: market analysis, user-needs mapping, competitive landscape, technical options, regulatory if applicable).
3. Wait for all subagents to complete.
4. Synthesize outputs into docs/matilha/research/<slug>-research.md with sections per subagent + a "Key findings" synthesis at the end.
5. Update project-status.md: current_phase: 10, phase_status: in_progress, next_action: "Run matilha-plan to write spec + plan".

Rules: Do

• Use parallel subagent dispatch (multiple Task tool calls in a single message).
• Cite sources in research output.
• Separate findings from recommendations.
• Update project-status on success only.

Rules: Don't

• Write the spec yourself (that's matilha-plan).
• Recommend decisions (research surfaces options, user decides).
• Advance current_phase past 10.

Expected Behavior

Output is a 500-1500 word markdown with structured sections. User reads it, forms a mental model, then invokes matilha-plan. If research surfaces blockers, log them in project-status.md:pending_decisions.

Quality Gates

• Research output exists and is non-empty.
• project-status.md shows current_phase: 10.
• pending_decisions populated if blockers found.

Companion Integration

• If ux-* skills from matilha-ux-pack are available: dispatch an additional ux-research subagent via Task tool for cognitive/perception considerations.
• If growth-* skills from matilha-growth-pack are available: dispatch a growth-research subagent mapping JTBD forces-of-progress.
• Otherwise: proceed with the 3-5 core subagents above.

Output Artifacts

• docs/matilha/research/<slug>-research.md
• Updated project-status.md

Example Constraint Language

• Use "must" for: every finding cites a source.
• Use "should" for: dispatch research subagents in parallel (single message with multiple Task calls).
• Use "may" for: skip competitive analysis if scope is internal-only.

Troubleshooting

• "research output is thin": Re-run with more specific sub-topics.
• "subagent dispatch fails": Check Task tool availability (Codex requires multi_agent = true in ~/.codex/config.toml).

CLI shortcut (optional)

If matilha CLI is installed (matilha --version succeeds), you can run matilha scout <slug> to execute this deterministically. The plugin path above works without any CLI installation.