PART 04 — PHASE 2

How to establish trust.

A source can provide evidence, never a safety guarantee. Review provenance, the pinned package and dependencies, required permissions, test evidence, ownership, monitoring, and removal before adoption.

10–12 Min Claude

Reviewed 31 August 2026. Anthropic's current Agent Skills overview defines a skill as a filesystem package that may include executable code. Model-assisted review can help find questions; it cannot certify a package or replace manual code review.

Part 1: Three provenance starting points

Start with provenance, then apply the same technical review to every source. Popularity, branding, and public visibility reduce some uncertainty but do not prevent compromise or unsafe updates.

START 1 · AUTHORITATIVE

Maintainer-controlled source

Confirm the exact publisher through current official documentation, then record the release or commit and integrity value. Publisher identity does not eliminate bugs, compromise, or environment mismatch.

Treat as: strongest provenance; still inspect and test.

START 2 · TRACEABLE

Inspectable public repository

History, issues, ownership, and diffs provide evidence. Stars and contributor counts are social signals, not controls; verify tags, dependencies, release artifacts, and maintainer changes.

Treat as: pin, inspect every file, and test.

START 3 · UNVERIFIED

Attachment or mutable link

Identity, integrity, and update history may be absent. Do not install it to inspect it and do not upload confidential contents to an unapproved reviewer.

Treat as: quarantine; obtain an authoritative source or reject.

Part 2: Evidence to record before approval

Do not rely on a visual badge or a changing install count. Keep a small review record so another person can reproduce the decision and identify what changed later.

✓
Origin
Authoritative URL, maintainer identity, license, and how identity was verified.
PROVENANCE
✓
Version
Immutable commit or release, package hash or signature where available, and review date.
INTEGRITY
✓
Reach
Every file, script, dependency, tool, network endpoint, readable path, and writable target.
LEAST PRIVILEGE
✓
Lifecycle
Owner, approver, isolated test evidence, monitoring, expiry, update review, rollback, and removal.
CONTROL
Trust is version-specific

Approval applies to the exact bytes and configuration reviewed—not to a package name forever. Pin versions or hashes where the distribution mechanism allows it, review every update, and remove approval when ownership, contents, permissions, or dependencies change unexpectedly.

Part 3: A model-assisted second pass

After a human inventories and reviews the complete package, a model can surface missed questions. Use an approved environment and sanitized content; never paste credentials, customer data, or proprietary code into an unapproved service.

Act as a second reviewer of this complete, version-pinned skill package. Do not label it safe. For every file: 1. Map instructions, scripts, dependencies, subprocesses, file reads/writes, tool calls, endpoints, and transmitted fields. 2. Flag dynamic downloads, unpinned dependencies, obfuscation, credential paths, hidden actions, approval bypass, destructive operations, and broad globs. 3. State the minimum permissions and synthetic fixtures needed to test it. 4. Identify uncertainties, missing files, generated artifacts, and claims you cannot verify. 5. Propose representative, edge, adversarial, failure, and rollback tests. Return evidence with file and line references. A human reviewer owns the decision. --- SANITIZED PACKAGE CONTENTS BELOW --- [include every text file and a manifest/hash list for all other files]

Verify each finding in the source. A clean model response is not due diligence: models can overlook behavior, misunderstand code, or hallucinate. Unknown binaries, missing dependencies, or code your reviewer cannot assess remain unresolved.

Part 4: The audit workflow, step by step

1

Quarantine and inventory the package without installing it

List every file, type, size, link, generated artifact, and archive member. Keep it outside project and personal skill directories; do not execute preview helpers.

2

Verify origin, version, integrity, and ownership

Trace the maintainer through an authoritative channel. Record an immutable revision and hash or signature where available; inspect release and ownership changes.

3

Review every file, dependency, and permission manually

Map inputs, outputs, scripts, subprocesses, tools, file paths, endpoints, data fields, credentials, destructive actions, and failure behavior. Get qualified review for unfamiliar code.

4

Use independent tools and model review as additional evidence

Scan dependencies and secrets, inspect diffs, and run the prompt above only with sanitized contents in an approved service. Resolve findings in source; never auto-approve a model verdict.

5

Test the pinned bytes with synthetic fixtures and minimum access

Use an isolated environment without secrets or production data. Require explicit approval for writes, messages, purchases, deployments, or other external effects. Capture logs, failures, and rollback; assign an owner and re-review date.

Re-review every update and periodically reassess unchanged packages as runtimes, dependencies, permissions, and threats evolve. If behavior deviates, disable or remove the skill, preserve logs, rotate any possibly exposed credentials, and follow the documented rollback or incident process.

← Part 3 — What a skill is Part 5 — Build your first skill →