The Agent Skills Directory

[SAFE]: The skill incorporates a 'Multi-Agent Safety Rule' and instructions for 'Safety Checks' to prevent the accidental staging of secrets, tokens, or code modified by other processes. These practices ensure that only intended, reviewed changes are committed to the repository.- [COMMAND_EXECUTION]: The skill uses local git CLI tools for repository management and instructs the agent to run project-defined verification scripts such as tests or linters. These operations are limited to the local environment and the current workspace context.- [SAFE]: Evaluation of the indirect prompt injection surface reveals that the skill ingests data from git status and git diff outputs in SKILL.md. Although the agent has the capability to execute shell commands for git and testing purposes, the workflow mitigates risk through granular file staging and mandatory manual review. No explicit boundary markers are used for text interpolation, but the skill includes instructions to sanitize output by checking for sensitive information like tokens before committing. (Ingestion points: git status and git diff outputs in SKILL.md; Boundary markers: Absent; Capability inventory: Shell execution for git and verification tools; Sanitization: Explicit secret and token scanning instructions).

commit-work