The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Documentation in references/commit-paths.md directs the agent to install additional skills using npx skills add a-green-hand-jack/ml-research-skills, which downloads code from a repository belonging to the author.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute shell commands, including Git operations and local Python scripts like scripts/validate_skills.py and scripts/prepare_sidecar_task.py for repository validation and maintenance.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted Git data that could contain malicious instructions.
Ingestion points: Repository metadata enters the context via git status, git log, git diff, and git show (documented in SKILL.md and references/commit-paths.md).
Boundary markers: Absent. The instructions do not define delimiters or provide warnings to the agent to ignore instructions embedded in the Git output.
Capability inventory: The agent has access to powerful tools including Bash, Write, Edit, and Glob, which could be exploited if an attacker embeds instructions in commit messages or file diffs.
Sanitization: Absent. There is no logic provided to sanitize or validate the content of the repository metadata before it is processed by the agent.

safe-git-ops