minhtran3124/agent-harness/feature-intake
- community
Checking that the SKILL.md file is still in the repository.
First gate for a change request: classify its risk lane and confidence, record the decision in `specs/<slug>/SUMMARY.md`, then route to the appropriate workflow. Use before research, planning, or edits.
Install
npx vskill@latest install minhtran3124/agent-harness/feature-intakeSKILL.md
Feature Intake — classify and route
<HARD-GATE> Do not edit, scaffold, or dispatch implementation until intake records Lane: and Confidence: in specs/<slug>/SUMMARY.md. The lane is risk; confidence is ambiguity. Never ask the user to classify risk; ask only to resolve a material ambiguity or narrow a hard-gated scope. </HARD-GATE>
Procedure
- Read
harness-manifest.jsonfor canonical hard-gate vocabulary and modes. Read
rules/orchestration.md and rules/auto-correct-scope.md for routing and branch policy.
- Classify the input: new spec, spec slice, change request, initiative, maintenance, or harness
improvement. Mark applicable risk flags: auth, authorization, data model, audit/security, external systems, public contracts, cross-platform, existing behavior, weak proof, and multi-domain.
- Assign the lane: a manifest hard gate is
high-risk; otherwise 0–1 flags istinyonly for
one-file/no-new-public-callable work (else normal), 2–3 is normal, and 4+ is high-risk. A human may lower a hard-gated lane only by narrowing scope.
- Assign confidence:
highfor one clear interpretation,mediumfor a safe documented
default, and low for materially different plausible interpretations. Low confidence stops for human confirmation regardless of lane.
- Write
SUMMARY.mdusingtemplates/SUMMARY.template.md, including the user intent verbatim:
``text Lane: <tiny | normal | high-risk> Confidence: <high | medium | low> Reason: <flags/hard gate or none> Flags: <comma-separated or none> Affects: <contract/module from PROJECT.md or none> Input-type: <type> Route: <path> Escalate: <yes (reason) | no> ``
- Initialize
runtime/run_state.pyasinvestigatingon a best-effort basis. For normal and
high-risk lanes, transition to planning; failures are observability failures, never an intake blocker. Run python3 scripts/verify_summary.py --lane <slug> before handoff.
Routes
| Lane | Route | |
|---|---|---|
| tiny | Create a branch, make the direct patch, and retain quick-check proof; report done only when that check passes, else state what is missing. | |
| normal | using-git-worktrees → subagent-driven-development. | |
| high-risk | brainstorming → xia2 → writing-plans → using-git-worktrees → subagent-driven-development; use compound for durable decisions. |
All lanes require branch isolation. Artifact requirements scale by signal: plan for more than three steps or two files, research for unfamiliar code or high-risk work, and a design for a real design fork or high-risk work. FULL_ARTIFACTS=1 forces all artifacts.
Research-depth handoff
When intake metadata exists, xia2 must use it: high-risk starts Deep, normal starts Standard, and tiny may be Quick only after every Quick condition in rules/research-depth.md passes. Xia2 does not re-score risk; it can only increase research depth when evidence reveals a deeper signal.
Arguments
$ARGUMENTS is the requested change. Derive <slug> using the ticket-prefix convention in templates/structure/specs-README.md; never rename an existing spec directory.
References
harness-manifest.json— hard gates and modes (authority)rules/orchestration.md— artifact and escalation policyrules/auto-correct-scope.md— branch and autonomy boundaries
Security
- Security scanPassed, 95 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/feature-intake/SKILL.md
Popularity
- Repository stars
- 7
- 7-day trend score
- 10
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/minhtran3124/agent-harness/feature-intake)