The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill represents an indirect prompt injection surface because it reads and processes data from log files and external articles to automatically generate and apply fixes to its own code and instructions.
Ingestion points: Reads from memory/logs/, memory/cron-state.json, and output/articles/ to identify failures and improvement areas.
Boundary markers: The skill does not define explicit delimiters or instructions to ignore malicious content embedded within the logs or articles.
Capability inventory: The skill can modify skill instructions (SKILL.md), agent prompts (CLAUDE.md), and configuration (aeon.yml), then commit these changes and open Pull Requests.
Sanitization: No sanitization process is described for the log data before it informs the generated code fixes.
[DYNAMIC_EXECUTION]: The skill performs self-modification of its own instructional code and configuration, which is subsequently executed during validation steps.
Evidence: Step 4 and 4b involve implementing fixes to skills and then executing bash scripts/dry-run.sh to run the modified skill in a dry-run environment.
[COMMAND_EXECUTION]: The skill utilizes local shell commands and specialized CLI tools to perform repository and development tasks.
Evidence: Uses the gh CLI (gh pr list, gh pr create) for GitHub operations and standard git commands (checkout, commit, add) for version control.

self-improve