The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local script node scripts/resolve-feature.js as part of its feature resolution logic to identify the relevant directory for requirements documentation.
[INDIRECT_PROMPT_INJECTION]: The skill retrieves data from external web sources via WebSearch, WebFetch, and agent-browser, which introduces a potential vector for indirect prompt injection. Mitigation includes mandatory instructions to ignore any embedded commands within fetched data.
Ingestion points: External data from web research tools and external documentation fetches.
Boundary markers: Directives to cross-verify claims and disregard instructions found in external sources.
Capability inventory: Write access to project directories, local command execution (node, git, grep), and agent delegation.
Sanitization: Automated 2-tier secret redaction scan applied to analyzed data.
[DYNAMIC_EXECUTION]: The skill dynamically constructs prompts and arguments for sub-agents and specialized skills like deep-research or codex-brainstorm based on the context discovered during analysis.
[PRIVILEGE_ESCALATION]: The skill implements a scope gate and path validation check to ensure that user-provided paths do not attempt directory traversal (e.g., using .. or absolute paths) outside the project scope.

req-analyze