The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands such as git, mkdir, cat, ls, chmod, and jq to manage repository files and configurations. It also generates a script that uses an inline python3 command to parse JSON data.
[DYNAMIC_EXECUTION]: The skill programmatically generates multiple shell scripts (check-docs.sh, check-skill-drift.sh, and check-md-lint.sh) using local repository content as input. These scripts are then registered in .claude/settings.json to be executed automatically by the agent runtime on specific tool calls.
[PRIVILEGE_ESCALATION]: The skill uses chmod +x to grant execution permissions to the shell scripts it generates, which is necessary for the agent to run them as hooks.
[INDIRECT_PROMPT_INJECTION]: The skill extracts content from documentation files (e.g., docs/code-structure.md, ARCHITECTURE.md) to create reminders. These reminders are echoed to stderr during file editing operations, which injects the extracted text into the agent's context. This creates a vulnerability surface where malicious or misleading instructions in the documentation could influence the agent's subsequent behavior.
Ingestion points: docs/code-structure.md, docs/code-guidelines.md, ARCHITECTURE.md, .flowchad/flows/.
Boundary markers: None present; reminder text is echoed directly to the agent's error stream.
Capability inventory: Execution of generated shell scripts, modification of agent configuration files, and use of chmod for permission management.
Sanitization: No sanitization is performed on the extracted documentation content before it is used in the reminder strings.

hookshot