The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the persistence of potentially untrusted data from a task session into permanent project instructions and new skill definitions. * Ingestion points: Information gathered during task execution from codebase content, conversation history, or external inputs as described in SKILL.md. * Boundary markers: Absent; the lesson-template.md does not provide delimiters or instructions to isolate or ignore embedded commands. * Capability inventory: File system write access to project-level instruction files (.cursorrules, llms.txt, CLAUDE.md) and the ability to create new executable assets in the skills/ directory. * Sanitization: Absent; no validation or filtering is performed on the codified lessons before they are written to the codebase.
[PROMPT_INJECTION]: The skill uses imperative language to modify the agent's operational cycle and override standard behavior. * Evidence: Directives such as 'Execute this meta-skill ALWAYS', 'You MUST trigger this skill', and 'Do NOT close a complex task without running this skill' in SKILL.md attempt to force behavior beyond standard completion workflows.

self-reflection