The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands (git diff, git log) using a reference point (commit, branch, or tag) provided directly by the user.
Evidence: Step 1 instructions: "Capture the diff command once: git diff <fixed-point>...HEAD... Whatever the user said is the fixed point... pass it through."
Risk: Direct interpolation of user input into shell commands can lead to command injection if the environment does not properly sanitize shell metacharacters.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data from the repository (code diffs and specifications) and passes it to sub-agents without boundary markers.
Ingestion points: Git diff outputs, commit messages, and the contents of files found in docs/, specs/, or .scratch/ (SKILL.md, Steps 1 & 2).
Boundary markers: Absent. The sub-agent prompts defined in Step 4 interpolate diff and spec content directly without delimiters or instructions to ignore embedded commands.
Capability inventory: The skill spawns 'general-purpose' sub-agents via the Agent tool, which may have significant capabilities depending on the host environment (SKILL.md, Step 4).
Sanitization: Absent. The skill does not perform validation or escaping of the content extracted from the repository prior to analysis.
[EXTERNAL_DOWNLOADS]: The skill references external setup routines and integrated issue-tracking workflows.
Evidence: Step 2 describes fetching issue data via a workflow in docs/agents/issue-tracker.md, and the main instructions suggest running a setup command /setup-matt-pocock-skills.
Note: These references likely point to project-specific automation for interacting with external platforms like GitHub or GitLab, or for configuring the local environment.

review