The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests 'plays'—instructions or rules—from multiple Markdown sources (playbook.md) including project-level files and shared folders. These plays are intended to be followed by the AI agent, creating a risk that malicious rules could be introduced into the project or personal configuration to influence agent behavior.
Ingestion points: Reads content from ~/.claude/aida/playbook.md, <projectPath>/playbook.md, and external directories specified in the project's configuration file (project.json) which are parsed in scripts/playbook-actions.sh.
Boundary markers: The system relies on Markdown headings (##, ###) and specific bold text patterns (**What:**, **Rationale:**) to parse and separate individual rules, though it does not implement strict sanitization against instructions embedded within the rule text itself.
Capability inventory: The skill can append new rules to files, create temporary files, and execute Git commits through a commit_project helper, providing a mechanism for persisting changes to the rulebooks.
Sanitization: Content is processed using jq for JSON data and awk with variable binding (-v) for Markdown manipulation, which helps prevent direct shell command injection during the rule capture and loading processes.

playbooks