The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/collect-integration-facts.py contains a function branch_policy() that uses exec() and compile() to dynamically load and execute code from a file located at a computed path (../../git-commit/scripts/branch_name.py). While used for cross-skill logic sharing, this pattern bypasses standard module importing and poses a risk if the file system structure is manipulated.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Git repositories which can be influenced by external actors (e.g., via commit messages or file content).
Ingestion points: The collect-integration-facts.py script reads file paths, status porcelain records, commit logs, and entire file contents for hashing purposes.
Boundary markers: The collected data is emitted as raw JSON without the use of explicit delimiters or instructions for the agent to ignore potentially malicious content embedded within the Git data.
Capability inventory: The skill has significant capabilities including executing arbitrary Git commands via subprocess.run, performing rebase/merge operations, and writing to the file system through temporary worktrees.
Sanitization: There is no evidence of sanitization or filtering of the repository-sourced data before it is presented to the agent's context.
[COMMAND_EXECUTION]: Both the Python script and the instructions in references/integration-workflow.md rely heavily on executing shell commands via subprocess.run and direct bash calls. These commands involve interpolating variables such as branch names and repository paths, which could lead to command injection if those inputs are not strictly validated by the calling agent.

worktree-rebase-merge