The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local git commands and a bash script (finalize.sh) to automate the creation and verification of feature branches. It also utilizes node as a JSON parser for temporary configuration data.
[DATA_EXFILTRATION]: The skill reads local repository files (autoresearch.jsonl, autoresearch.md, autoresearch.ideas.md) to extract metrics and context for branch documentation. This access is localized to the repository and does not involve external network transmission.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it processes external files into the agent's context.
Ingestion points: Reads experiment notes and ideas from autoresearch.md and autoresearch.ideas.md in Step 1 and Step 3.
Boundary markers: The skill lacks explicit delimiters for file content, though it includes a human approval checkpoint before the execution of the finalization script.
Capability inventory: The skill can execute shell commands, create git branches, and modify repository state via file checkouts.
Sanitization: Filenames and commit metadata are handled with appropriate quoting in shell commands, but natural language content is not filtered for embedded instructions.

autoresearch-finalize