The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill provides numerous templates for creating shell and TypeScript scripts that are executed at runtime as hooks. Several bash templates in references/hook-templates.md and references/visibility-patterns.md utilize eval echo "$FILE_PATH" to expand home directories, which represents a potential command injection vector if the tool-provided file path is maliciously manipulated.
[COMMAND_EXECUTION]: The instructions and references (e.g., SKILL.md, references/debugging-guide.md) guide users to execute shell commands and modify file permissions using chmod +x to enable hook execution. This is consistent with the skill's purpose but involves granting execution privileges to local scripts.
[INDIRECT_PROMPT_INJECTION]: The skill documents an attack surface for indirect prompt injection by providing templates that ingest untrusted tool data via PAYLOAD=$(cat) and interpolate it into feedback for the agent. 1. Ingestion points: Tool input and output data are read from stdin in templates found in references/hook-templates.md and SKILL.md. 2. Boundary markers: The templates do not include boundary markers or instructions to ignore embedded commands within the processed data. 3. Capability inventory: The documented hooks have the capability to execute arbitrary commands, write to the filesystem, and influence the agent's decision-making loop as described in references/lifecycle-reference.md. 4. Sanitization: There is no evidence of sanitization or validation of the ingested tool data before it is returned to the agent context.

hooks-development