The Agent Skills Directory

[PROMPT_INJECTION]: The skill establishes a learning loop that ingests data from user feedback and command errors, which is then used as instructional context for future sessions.
Ingestion points: Writes to and reads from .learnings/LEARNINGS.md, .learnings/ERRORS.md, and .learnings/FEATURE_REQUESTS.md.
Boundary markers: Data is organized via Markdown headers but lacks explicit boundary markers or 'ignore' instructions to prevent the agent from executing malicious commands stored in the logs.
Capability inventory: The skill provides shell scripts for file creation (extract-skill.sh) and JS/TS handlers that inject content into the agent's bootstrap cycle.
Sanitization: No validation or escaping is applied to user-provided corrections or error logs before they are written to the persistent store.
[EXTERNAL_DOWNLOADS]: The documentation references installation from external repositories.
Evidence: SKILL.md contains instructions for git clone from github.com/peterskoett/self-improving-agent.git and use of the clawdhub installer.
Context: These references are for the installation of the skill's own source code and target a well-known service (GitHub).

self-improvement