The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from external sources, including GitHub issues (via the -f #N flag and gh tool), web research results (via subagents), and arbitrary local files (via -f ), creating an indirect prompt injection surface.\n
Ingestion points: step-01-analyze.md facilitates loading requirements and context from external files and issue trackers.\n
Boundary markers: The SKILL.md and step-01-analyze.md files include a 'Trust model' section with explicit instructions to treat source content as data and ignore any embedded directives.\n
Capability inventory: The skill uses tools for file writing (TodoWrite), command execution (git, bash), and network access via subagents.\n
Sanitization: Instructions are provided to the model to strictly treat fetched content as evidence rather than directives.\n- [COMMAND_EXECUTION]: The skill relies on local bash scripts to handle state transitions, template initialization, and progress tracking.\n
Evidence: Files such as setup-templates.sh, update-progress.sh, resume_lookup.sh, and validate_state.sh are executed by the agent to manage the workspace under ~/.agents/output/. These scripts utilize standard system utilities like awk, grep, and sed to perform their tasks securely.

apex