The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of system commands for repository management and workflow execution. It uses gh pr list and gh pr create for GitHub interaction, and git checkout, git add, and git commit to manage changes. It also executes a local test script via bash scripts/dry-run.sh.
[DYNAMIC_EXECUTION]: The skill is designed to modify the agent's own code base, specifically other SKILL.md files and CLAUDE.md instructions. After performing these modifications, it executes the updated code using the dry-run.sh script to verify the changes. This creates a self-modification and execution loop.
[INDIRECT_PROMPT_INJECTION]: The skill contains a vulnerability surface where instructions from processed data could influence the agent's behavior.
Ingestion points: The skill reads memory/logs/, memory/MEMORY.md, and output/articles/. These files contain a history of the agent's past actions and tool outputs, which may contain untrusted data from external interactions (e.g., website content, API responses recorded in logs).
Boundary markers: There are no explicit instructions or delimiters defined to isolate the content of these logs from the agent's primary instructions.
Capability inventory: The skill possesses the ability to write to the file system, commit changes to the repository, create pull requests on GitHub, and execute shell scripts.
Sanitization: The skill lacks a mechanism to sanitize or validate the content retrieved from logs before using it to generate and apply code fixes.

self-improve