The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various local shell utilities (e.g., find, awk, jq, stat, chmod) to perform its diagnostic and tracing functions. These operations are restricted to the local filesystem and do not involve network activity.
[DYNAMIC_EXECUTION]: The skill-trace.sh script generates a bash-based canary block from a template and injects it into other SKILL.md files. This block is intended to be executed by the agent when the traced skill is activated. The generation process uses shell_quote to ensure paths are safely escaped, mitigating command injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface where it ingests untrusted data from other skill files (e.g., SKILL.md). Ingestion occurs in skill-trace.sh and skill-probe.sh. The skill implements boundary markers and uses sanitization methods like shell_quote and sr_json_escape when interpolating data into commands or logs. Its capabilities include subprocess calls (e.g., find, awk) and file writes, but these are managed with strict local permissions.

skill-debug