The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/tracker_template.py script executes various git commands via subprocess.run to manage repository state, fetch updates, and inspect commit history. Arguments are passed as a list, mitigating shell injection risks.
[EXTERNAL_DOWNLOADS]: The tracker script performs git fetch operations from remote repositories hosted on GitHub. Documentation in SKILL.md also provides instructions for downloading patch files using curl.
[REMOTE_CODE_EXECUTION]: The workflow described in SKILL.md includes downloading and applying patches directly from GitHub using curl and git apply. While this modifies the local codebase with external content, the instructions emphasize manual auditing and verification steps before merging.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from untrusted upstream repositories, creating a potential surface for indirect prompt injection.
Ingestion points: Commit messages, titles, and file diffs are retrieved from the upstream repository via git log and git diff-tree (in tracker_template.py) and presented to the agent for auditing.
Boundary markers: The skill does not implement explicit delimiters or instruction-ignore warnings when presenting upstream data to the agent context.
Capability inventory: The agent is granted the capability to apply code changes via git apply or git cherry-pick and to execute local test suites based on its evaluation of the upstream data.
Sanitization: The retrieved upstream content is not sanitized or filtered for instructions before being evaluated by the agent.

fork-upstream-sync