The Agent Skills Directory

[PERSISTENCE]: The skill modifies global agent configuration files, including ~/AGENTS.md, ~/.claude/CLAUDE.md, and ~/.codex/AGENTS.md. These modifications are designed to persist agent instructions and rules across all future projects and sessions in the user's environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data by reading from and writing to shared files such as agent-docs/ledger.md and agent-docs/lessons/*.md. It appends task evidence and promotes lessons into instructions without explicit sanitization or boundary markers to prevent the agent from obeying instructions embedded in the logs.
[COMMAND_EXECUTION]: The instructions require the agent to write and execute regression tests as part of the post-task analysis. This involves the dynamic generation and execution of code based on the context of the analyzed failure.

agent-rsi