The Agent Skills Directory

[SAFE]: The skill is designed with a strong focus on security and privacy. It mandates the use of isolated 'lab' environments (mktemp -d) and isolated application profiles (e.g., --user-data-dir, --use-mock-keychain) to ensure that the user's primary profile and sensitive data remain untouched during modification tasks.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because its primary function involves ingesting and analyzing third-party code, manifests, and runtime data. It mitigates this risk through a mandatory 'Publish Check' using scripts/publish_check.py, which scans for secrets and sensitive metadata, and a requirement for human review of all artifacts before they are moved to the final ship tree.
Ingestion points: Reads code, manifests, and captured logs from targeted applications (documented in SKILL.md Step 4 and Step 5).
Boundary markers: Incorporates verification steps such as modlog.py lint-note and the publish_check.py filter to validate evidence paths and contents.
Capability inventory: Uses subprocess calls to interact with target application CLIs, system tools like codesign, and security tools like gitleaks (found in all provided Python scripts).
Sanitization: Employs scripts/publish_check.py to filter out credentials, personal paths, and hardware identifiers using extensive regular expression patterns and gitleaks integration.
[COMMAND_EXECUTION]: The skill executes shell commands through Python's subprocess module to drive legitimate tools like gitleaks for secret scanning, codesign for checking application signatures, and the target application's own CLI tools. These operations are restricted to the context of the skill's defined safety rules and isolated lab environments, with instructions to prioritize sanctioned APIs and avoid dangerous patterns like pkill.

mod-anything