The Agent Skills Directory

[SAFE]: No malicious patterns detected. The skill uses standard local development tools (wc, grep, claude plugin validate) for technical validation of word counts, file structures, and manifest formats. No network operations or credential harvesting patterns were identified.
[COMMAND_EXECUTION]: The skill utilizes local shell commands for quality assurance tasks, such as validating word counts in SKILL.md and verifying plugin manifests. These operations are restricted to the local development environment and serve legitimate developer workflows. A Python snippet for counting headings is provided as a local verification example and uses only standard libraries (pathlib, re).
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as its primary function is processing 'build prompts' and 'donor' files to generate output. 1. Ingestion points: AGENT_REPO/prompts/*.md, USE-CASES.md, and donor files referenced in provenance.md. 2. Boundary markers: The methodology focuses on structural integration rather than explicit isolation of untrusted input. 3. Capability inventory: Includes local file read/write operations and execution of diagnostic CLI tools. 4. Sanitization: Relies on manual developer review gates (Phase A3, A7) to validate generated content.

forge