| name | repo-discovery |
| description | Phase-1 repo reconnaissance as a standalone skill: stack detection, relevant files, validation commands from repo evidence, workflow hints, and risk flags. Use inside /scout, /ks-conductor pre-flight, or any read-only discovery before planning or implementation. |
On-demand loading: Read this skill only when the task clearly matches the description above. Do not load for unrelated work.
Repo discovery
Purpose
Execute minimal effective reconnaissance to map the real implementation surface: stack, boundaries, commands, tests, and risks—without editing files or designing the solution.
When to use
- Start of
/scout Phase 1 (read this skill and align output with scout report format).
Not for: Full onboarding doc for new team members — use codebase-onboarding (parallel explore + synthesized guide).
/ks-conductor when skipping a full /scout invocation but discovery is still required.
/build-full pre-flight when the execution pack lacks verified commands or file paths.
When to skip
- User gave exact file paths, symbols, and a single known fix with no ambiguity.
- Another agent just produced a fresh
scout-YYYY-MM-DD-*.md for the same task.
Procedure
- Scope: Restate the task in one line; note workspace root vs dotfiles-only (no project repo).
- Stack: From manifests (
package.json, pyproject.toml, Gemfile, go.mod, etc.), lockfiles, CI, and directory layout—infer framework, package manager, monorepo layout.
- Entry points: App entry, API routes, main packages, env/config patterns (names only—do not read secrets).
- Relevant files: List files/modules likely touched; note imports / imported-by when cheap to determine.
- Commands (verified only): Extract install, lint, typecheck, test, build, dev from manifests, Makefile, CI workflows—never invent script names.
- Constraints:
.cursor/rules, /docs/, migrations folder, security-sensitive areas (auth, payments, uploads).
- Tests: Where tests live; unit vs e2e; gaps if obvious.
- Risks: Contract changes, migrations, auth, performance, duplication.
Output shape
Produce bullets or tables suitable for pasting into:
- The Scout report sections (stack, workflow type, files, validation commands, risks).
- YAML front-matter fields:
classification, confidence, workflow_type, next_command (see phase-handoff / /scout artifact schema).
Integration
- After discovery,
/scout must still run Phase 0 (user-devcontext, optional brainstorming) and Phase 2 (user-Sequential Thinking when triggered) unless this skill is invoked only as a sub-step inside scout.
- Prefer subagent
explore for large unknown repos when parallel speed helps.