The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill implements a self-modification pattern. The 'Self-Evolving Skill' and 'Post-Execution Reflection' sections explicitly instruct the agent to use the Edit tool to update the SKILL.md file itself based on runtime failures or successes. This allows the skill's core logic and instructions to be altered permanently by the agent.
[INDIRECT_PROMPT_INJECTION]: The skill operates on untrusted data from the local filesystem that is processed into the agent's context.
Ingestion points: The output of chezmoi status and chezmoi diff --no-pager is read by the agent to detect drift and present choices to the user.
Boundary markers: None. The raw shell output from the diff command is processed directly without delimiters or 'ignore' instructions.
Capability inventory: The skill has access to Bash (arbitrary shell command execution), Edit (ability to modify any file, including the skill's own instructions), and AskUserQuestion (interaction with the user).
Sanitization: No sanitization or escaping is performed on the file content before it is processed by the LLM.
Risk: If a file on the system contains malicious prompt injection strings, they will appear in the diff output. Under the 'Self-Evolving' instructions, the agent might be misled into incorporating those malicious instructions into the SKILL.md file.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute chezmoi and git commands. While these are intended for dotfile synchronization, they operate on arbitrary file paths and execute shell pipelines involving re-add, commit, and push operations.

chezmoi-sync