Install
$ agentstack add skill-dylanmccavitt-agent-skills-scout ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
Scout
Turn an unclear task into decisions the user actually made.
Session protocol
- Fixed sweep, in order:
scope → shape → data → edges → seams → done-looks-like. A territory closes ✓ grilled or ⊘ skipped with a one-line reason; the session ends only when all six are closed. - One question per turn: ≤2 lines, 2–3 mutually exclusive options, recommendation first with its tradeoff, asked through the host's native input control.
- Explore instead of asking when the codebase answers it.
- After every answer print the ledger, then the decision as one line:
✓N decided · ● · ○
- "I don't know" never stalls: it becomes a named spike in the brief.
Advisor
At each territory border, a subagent reviews for missed questions and wrong assumptions. It may inject at most 2 bonus turns (marked [advisor]) or clear the border silently.
Session trace
Hosts read the session as JSONL. Print each beat as one unindented line:
- each turn, the ask then its ledger:
{"type":"ledger","turn":1,"text":"✓1 decided · ● scope · ○ shape, data, edges, seams, done-looks-like"} - each border, in sweep order:
{"type":"advisor_review","territory":"scope","bonus_turns":0} - at the end, sweep (all six, in order) then brief:
{"type":"scout_sweep","territories":[{"name":"scope","status":"grilled"},{"name":"shape","status":"skipped","reason":"settled earlier"}]}then{"type":"decision_brief","decisions":["..."],"spikes":["..."],"now":"...","next":"...","later":"..."}
Where artifacts exist, the same beats drive one board.
Exit brief
Fixed shape, compact: numbered decisions (one line each) · spikes · Now / Next / Later. Deliver it in chat and on the board, then record it once: decision-shelf new "" --status selected fills the record and set status selected in one call. End with Record: as the last line.
Evolve the plan
After selection, a material change starts with decision-shelf propose; the current revision stays active until checkpoint folds it or reject closes it, and checkpointing needs matching authority.
Soft enforcement
Implementation that deviates from a scouted decision says so in one line before the code — then proceeds. No gates, no freeze, no check commands.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: dylanmccavitt
- Source: dylanmccavitt/agent-skills
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.