The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes several automation scripts (Python and Shell) for managing the learning store. Specifically, scripts/run_in_wsl.ps1 serves as a bridge for Windows users to execute the project's shell scripts via WSL, involving the construction and execution of shell command strings.
[PROMPT_INJECTION]: The skill facilitates the modification of core agent instruction files (e.g., AGENTS.md, CLAUDE.md, .github/copilot-instructions.md) by 'promoting' captured learnings into these files. This creates an indirect prompt injection surface if malicious or misleading instructions are logged and later promoted.
Ingestion points: Data is ingested from the .learnings/ directory, specifically from LEARNINGS.md, ERRORS.md, and FEATURE_REQUESTS.md (e.g., in scripts/promote_learning.py).
Boundary markers: While the skill uses specific header sections (e.g., 'Self-Improvement Promotions') to organize promoted rules, it does not implement programmatic wrappers or instructions for the agent to ignore potentially malicious embedded content within those rules.
Capability inventory: The skill has the capability to write to local instruction files via scripts/promote_learning.py and execute various shell commands through the included helper scripts.
Sanitization: No programmatic sanitization or validation of the learning content is performed before it is distilled and appended to the project's primary instruction files.

self-improvement