The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The scripts/validate-hooks-json.py script provides an interface for ingesting and validating user-defined JSON configuration files. While this is the intended functionality of a linting tool, it represents a potential surface for indirect prompt injection if an agent were to interpret data from a malicious configuration file during the validation process.
Ingestion points: The validate-hooks-json.py script reads JSON data from local files (e.g., hooks.json) specified by the user or detected in the repository.
Boundary markers: The script is a standalone validator that parses JSON structurally; it does not interpolate external content into the agent's system prompt.
Capability inventory: The script utilizes subprocess.run to invoke the git binary for path resolution and standard file I/O to read configuration files.
Sanitization: The script implements strict schema validation, checking keys against a pre-defined catalog of 30 known events and verifying data types using json.load() and explicit conditional checks.
[COMMAND_EXECUTION]: The skill includes a utility script, scripts/validate-hooks-json.py, which executes a local system command.
Evidence: The script uses subprocess.run(['git', 'rev-parse', '--show-toplevel']) to resolve the repository root for finding default configuration paths. This is a standard and benign use of subprocess for environment-aware tooling.
[SAFE]: The documentation within references/ includes external links to official Claude Code documentation and standard developer resources (e.g., GitHub, agentskills.io). These references are informative and originate from trusted or well-known technical domains.
[SAFE]: The skill documents the !command (Dynamic Context Injection) platform feature in references/skills-reference.md. As this is provided for educational purposes within a reference guide and is not used to execute hidden or malicious commands within the skill's own SKILL.md, it does not pose a security risk.

claude-code-ops