Skip to content
Back to skills

Scout

ASecurity

Investigate a defect with graphify and code references, then certify the root cause with runtime proof. Never modifies code. Runs standalone on `/scout <bug>`, or as the defect arm of `/recon`.

  • 3 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added September 5, 2026
ai-agentsgogit

Security analysis

A100/100

Pro scans all 7 files and shows the line behind each finding

Scanned September 5, 2026

npx -y skills add bravros/bravros --skill scout --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Scout?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Scout
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/bravros-scout/badge)](https://www.skillsdirectory.com/skills/bravros-scout)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: scout
description: Investigate a defect with graphify and code references, then certify the root cause with runtime proof. Never modifies code. Runs standalone on `/scout <bug>`, or as the defect arm of `/recon`.
---

# Scout — investigate, verify, certify

Read [briefing.md](references/briefing.md) on demand for detailed context and instructions.

INTENT: prove a root cause with runtime evidence, then hand the fix elsewhere. Investigation produces a hypothesis; only certification makes it a diagnosis. Findings land in `.planning/scout/S-NNNN-<slug>/`. Nothing certifiable → say so plainly (`UNCERTIFIED`), never ship a guess.

HARD CONSTRAINTS:
- **NEVER modifies application code.** Writes findings and reports only inside reserved `$SCOUT_DIR`; the fix happens through `/recon` → `/orchestrate` or `/quick`. Read-only covers *code*, not runtime *inspection* — reads, `SELECT`s and existing tests are fine; data writes, migrations and side-effecting jobs are not.
- **Certification is the gate.** A diagnosis needs reproduce + state match, a counterfactual, or an unbroken evidence chain (`references/certification.md`). Cap 3 rounds; then report `UNCERTIFIED` honestly and recommend deeper investigation, not a fix.
- **graphify is a lead source, never a verdict** — a confident graph hit still goes through verification; source code always wins.
- **The hand-off is the operator's decision** — always `ask_question`.
- **Called by `/recon`?** Skip the routing question and return `$SCOUT_DIR` plus the verdict; `/recon` folds `diagnosis.md` into its dossier as `01-diagnosis.md` and carries the certified cause into Traps and Closed decisions.

## Flow

1. Materialize engine: `mkdir -p .bravros/workflows && cp -f ~/.bravros/skills/scout/scripts/scout-investigate.js .bravros/workflows/scout-investigate.js`
2. Reserve dir: `SCOUT_DIR=$(bravros nextid reserve scout --slug "$SLUG"); SCOUT_ID=$(basename "$SCOUT_DIR" | grep -oE 'S-[0-9]+')`.
3. Build candidate lead list (`graphify`/`grep`/`git`/`error`).
4. Run parallel engine:
   ```
   Workflow({ name: 'scout-investigate', args: { scout_dir: SCOUT_DIR, bug: ARGUMENTS, category, stack, repro?, leads, boost, max_rounds: 3 } })
   ```
5. Write `diagnosis.md` + `report.md` in `$SCOUT_DIR` (schemas: `references/report-template.md`). Commit: `bravros commit "🔍 scout: $SCOUT_ID investigation for $SLUG" <files>`.
6. Route via `ask_question` (decision matrix: `references/investigation-guide.md`) — **standalone only**. Backlog route = write `B-NNNN` per `.planning/CONVENTIONS.md` (`bravros nextid` for ID). Invoked by `/recon`: return `$SCOUT_DIR` and the verdict instead, and let `/recon` own the routing.

Close investigation by appending `completed`/`cancelled` event to `.planning/events.jsonl`. Use `$ARGUMENTS` as bug description.

Files in this skill

  • SKILL.md2.8 KB
  • references/briefing.md2.9 KB
  • references/certification.md3.4 KB
  • references/investigation-guide.md3 KB
  • references/report-template.md6.9 KB
  • scripts/scout-investigate.js17.1 KB
  • skill.yaml2.1 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.