The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes git history, which serves as an untrusted data source that could contain malicious instructions.
Ingestion points: Data is ingested via git log, git show, and git blame as described in SKILL.md and reference/framework-templates.md.
Boundary markers: The LLM Fix Prompt template includes verbatim diff hunks and commit messages without explicit delimiters or instructions to disregard embedded content.
Capability inventory: The agent can execute git commands and arbitrary test scripts through the bisect run feature.
Sanitization: No specific sanitization or filtering of commit content is mandated before interpolation into prompts.
[COMMAND_EXECUTION]: The skill's core functionality relies on executing shell commands for git operations and performance profiling.
Evidence: The bisect, regression, and flame subcommands in SKILL.md involve subprocess execution of git and external profilers.
Safety: The instructions explicitly require the agent to obtain user confirmation before running test commands to mitigate arbitrary code execution risks.

trail