The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute repository-level commands for context gathering. This includes using git log, git diff, git stash, and grep to audit recent changes and existing code patterns. It also invokes the GitHub CLI (gh) to retrieve pull request metadata and repository defaults.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, potentially untrusted content such as plan documents, pull request descriptions, and commit messages. This creates a surface for indirect prompt injection attacks.
Ingestion points: Data is ingested via the Read tool (reading TODOS.md, CLAUDE.md, and design docs), git diff outputs, and gh pr view JSON responses.
Boundary markers: The skill does not define explicit delimiters or instructions to ignore embedded commands within the ingested plan documents, relying on the agent's general instructions to process the text as a "plan under review."
Capability inventory: The skill possesses the capability to execute shell commands (Bash), perform network queries (WebSearch), and interact with the user (AskUserQuestion).
Sanitization: There is no evidence of input sanitization or filtering applied to the contents of the files or git metadata before they are incorporated into the agent's context for analysis.
[EXTERNAL_DOWNLOADS]: The skill uses the WebSearch tool to perform a "Landscape Check," querying for product categories, alternatives, and industry trends. While this is a network operation targeting external search engines, it is used strictly for information gathering relative to the plan review task.

plan-deep-review