agentsclimarketplace

Tool review gate

Skill wde123sadw/agent-preflight/skills/tool-review-gate

Review a proposed skill, plugin, MCP server, CLI, package, API, connector, application, or service before installation or use. Use when a new capability may execute code, access files or accounts, receive project data, require secrets, cost money, change configuration, or create external effects. Present a decision-ready approval card and wait for explicit, scoped user approval.From its SKILL.md

Install
npx -y skills add wde123sadw/agent-preflight --skill tool-review-gate

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 26 days oldThe repository was created 26 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.8 KB, 686 tokens by cl100k_base, as published. Nobody here has run it

Tool Review Gate

Goal

Convert candidate information into a decision the user can safely approve, reject, or constrain. Discovery is not authorization.

Establish the action under review

Name the exact:

  • candidate and canonical source;
  • version, commit, release, image, or package identifier when available;
  • installation or enablement scope;
  • requested permissions and credentials;
  • data it will receive and where data travels;
  • paid resources, quotas, or subscriptions;
  • configuration and external effects;
  • verification and rollback plan.

If the action is vague, do not request approval yet.

Perform the review

Read review-checklist.md. Verify claims using current primary sources and local read-only checks.

Rate each area independently as LOW, MEDIUM, HIGH, or UNKNOWN:

  • source and supply chain;
  • code execution and install scope;
  • filesystem, network, account, and system permissions;
  • secrets and authentication;
  • data exposure, retention, and residency;
  • financial and quota exposure;
  • maintenance, compatibility, and vendor lock-in;
  • reversibility and uninstall quality;
  • operational blast radius.

Do not collapse UNKNOWN into low risk. Do not claim a complete security audit unless one was actually performed.

Prefer the least-powerful option

Before recommending approval, consider:

  • reuse of an existing tool;
  • a project-scoped install instead of user or system scope;
  • read-only mode before write access;
  • dry-run, preview, sandbox, test account, or sample data;
  • narrower API scopes;
  • a local tool instead of external data transfer;
  • a one-time manual step instead of a persistent integration;
  • a pinned version and reproducible source.

If a lower-risk option still meets the requirement, recommend it.

Present the approval card

Use approval-card.md. Keep facts, unknowns, inference, and recommendation distinguishable.

Offer explicit choices:

  1. approve the exact proposed action;
  2. approve with a narrower scope or condition;
  3. use the no-install fallback;
  4. request more evidence;
  5. reject.

Wait for scoped approval

Require an explicit choice before:

  • installation, enablement, or update;
  • authentication or credential entry;
  • modification of global, shared, or production configuration;
  • new external data transfer;
  • payment, subscription, quota consumption, publication, deployment, deletion, or migration.

Approval must match the reviewed candidate and scope. A user saying “find a tool” is not approval. A user approving a local read-only test is not approval for account-wide write access.

Record an approval receipt before execution: the approved candidate or action, exact scope, constraints, environment, and any expiration or one-time limit. Treat the user's actual words as evidence; do not widen them through paraphrase.

Execute safely after approval

When another workflow performs the approved action:

  1. verify that the command and scope still match the approval card;
  2. prefer dry-run or health check first;
  3. capture version and installation evidence;
  4. verify only the required capability;
  5. avoid unrelated setup or optional telemetry;
  6. report rollback or removal instructions;
  7. stop if permissions, data flow, price, or source differs from the approved facts.

This skill reviews and gates. It does not silently execute the installation itself.

What ships with it: 3 files

3.4 KB alongside SKILL.md

agents/

Keep looking

Skills are one crate of 326,782. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.