The Agent Skills Directory

[COMMAND_EXECUTION]: The skill contains logic to discover and execute local CLI tools (claude, claude-code, codex, codex-cli, opencode, pi) using command -v. It also allows for the execution of arbitrary user-recommended agent commands. While the skill requires explicit user approval before execution, this creates a potential vector for local command execution if the user approves a malicious or unexpected path.
[INDIRECT_PROMPT_INJECTION]: The skill's workflow involves ingesting and triaging data from external reviewers (subagents or CLI fallbacks), which represents a vulnerability surface for indirect prompt injection.
Ingestion points: Reviewer summaries, blocking findings, and evidence recorded in the /vs_review/ Markdown reports (see references/review-report-template.md).
Boundary markers: The skill enforces 'fresh sessions' and limits context sharing to neutral 'navigation packets' to isolate the reviewer's reasoning from the main agent's history.
Capability inventory: The agent possesses the capability to write files, invoke subagents, and execute shell commands.
Sanitization: The skill emphasizes the use of a 'neutral navigation packet' for input but does not define explicit sanitization or filtering mechanisms for the reviewer outputs before they are processed by the main agent for triage.
[DYNAMIC_EXECUTION]: The skill dynamically constructs and runs shell commands for external CLI substitutes based on the results of local discovery or user input.
Evidence: The fallback logic described in SKILL.md (Section 5) and the Reviewer Launch Records in the report template demonstrate the dynamic assembly of executable commands.

subagent-vs-review