The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill clones external GitHub repositories to a local directory (../repo-name) for analysis. This is a core component of the audit and harvest modes described in SKILL.md and references/test-harvest.md.- [COMMAND_EXECUTION]: The skill uses various shell commands to process repository data: 1. Invokes git for repository management, diffing, and metadata retrieval (cloning, git rev-parse, git diff). 2. Uses rg (ripgrep) to perform exhaustive file searches and build inventories of tests and concepts. 3. Uses gh or gitcrawl to retrieve issue and pull request metadata from GitHub. 4. Executes a bundled Node.js script (scripts/validate-concept-matrix.mjs) to validate research ledgers.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted data from external sources, creating a potential vector for malicious instructions to influence the agent. 1. Ingestion points: External source code, issue threads, and pull request descriptions from cloned repositories (references/test-harvest.md). 2. Boundary markers: There are no explicit instructions or delimiters defined to separate untrusted external content from the agent's primary instructions during analysis. 3. Capability inventory: The agent possesses file system write access (docs/, .tmp/) and shell command execution capabilities, which could be leveraged if an injection is successful. 4. Sanitization: The skill includes a sanitization step for repository keys using tr to prevent path traversal or command injection in filenames. It also recommends clustering security-sensitive issues (CVEs, GHSAs) into a 'security-quarantine' to prevent them from being processed through normal harvest routines.- [DYNAMIC_EXECUTION]: The skill executes a local JavaScript utility (scripts/validate-concept-matrix.mjs) using node to validate research artifacts. The instructions guide the agent to invoke this script via shell commands provided in references/editor-audit.md.

research