The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The instructions direct the agent to run a local script scripts/detect_phases.sh. As this file is not provided within the skill package, its contents and security implications cannot be verified.
[COMMAND_EXECUTION]: The workflow for finding regressions utilizes git bisect run <test-command>, which allows the agent to execute arbitrary commands provided by the user or found in the environment.
[COMMAND_EXECUTION]: The skill makes extensive use of powerful Git operations including git reset --hard, git checkout, and git revert. These commands can cause data loss or modify the project history, although the skill includes some safety measures like auto-stashing and user confirmation steps.
[PROMPT_INJECTION]: The skill analyzes Git commit messages to automatically detect and label development phases. Because commit messages are untrusted data (particularly in public or collaborative repositories), they represent an indirect prompt injection surface where malicious instructions could attempt to influence the agent's phase identification or decision-making process.
Ingestion points: Git commit history (git log), tags, and branch names (SKILL.md).
Boundary markers: None specified for commit message parsing.
Capability inventory: Shell command execution via git bisect run, file system modification via git reset/checkout, and script execution (SKILL.md).
Sanitization: No explicit sanitization or filtering of commit message content is described.

git-phase-restore