tooling-scout-capability

name	tooling-scout-capability
description	Inbound half of the scout pair. Capability gap analysis for skills and MCP servers - what to ADD. Sweeps repos, hydrates against skillsmp and glama, security-audits silver/gold tiers, installs one-issue-one-commit. Paired with tooling-scout-displacement (what to SHED).

scout-capability

The inbound half of the scout pair. This skill answers "what capability is missing that I should acquire?" - external catalog in, matched to your repos. Its sibling tooling-scout-displacement runs the opposite direction: your custom code in, matched to off-the-shelf OSS, answering "what custom thing should I retire because something already does it?" Same scouting instinct, inverted flow. See Paired with below.

Triggers

capability scout, scout capability, find me skills, find me mcps, gap analysis, missing capabilities.

Six-phase routine. Each phase runs independently and checkpoints to a single-day vault inbox file, so the user can run "scout-capability phase 3" from the train and resume without re-running phase 1.

Why six phases: the full pipeline is too large for one model run, and each phase has a different cost/risk profile. Phase 1 is open-ended ideation, phase 4 is a security gate, phase 6 mutates the user's personal-OS repo. Mixing them into one invocation either blows context or makes the security gate easier to skip. Splitting forces an explicit checkpoint between "speculate" and "install."

Outputs go to the notes/scratch location, not the personal-OS repo. Speculative discovery incidentally surfaces private repo intent and personal context. Only the final per-install commits land in the personal-OS repo. Inbox path: <notes-dir>/YYYY-MM-DD-scout-capability-{phase}.md (parameterize to the user's scratch/notes location).

Phases

• Phase 1 - Grounded sweep + speculative ideation - walk the working surface, build a candidate list, brainstorm what doesn't exist yet.
• Phase 2 - Hydration - resolve bare names against skillsmp/glama plus an awesome-list backstop, dedup against installs.
• Phase 3 - Categorize and rank - semantic categories, global 3:2:1 medal ranking.
• Phase 4 - Security audit - 🥈/🥇 only, supply-chain-audit skill, 🟢🟡🔴 safety prefix (blocked on execution verification).
• Phase 5 - Present 🥇🟢 inline - flatten the top tier to chat, explicit-deny approval.
• Phase 6 - Install approved entries - one issue + one commit + one push per entry, defense-in-depth re-check.

Paired with

tooling-scout-displacement - the outbound half. The two scouts are one portfolio-management activity run in opposite directions:

• scout-capability (this skill) - catalog in, acquire. Finds capability you lack and should add. The input is the world's skills/MCPs.
• scout-displacement - inventory in, retire. Finds custom code you maintain that off-the-shelf OSS already does, and proposes shedding it. The input is your own subsystems.

Run scout-capability when the tooling surface feels too small. Run scout-displacement when it feels too big or too custom. They share the back half (research-a-candidate, security-audit, one-issue-one-commit landing) and differ only in what feeds phase 1.

See also

• Scrub-on-reject and notes - reject ordering, run cadence, resume model, ward wrapper paths, why speculative entries matter.