The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests Architecture Decision Record (ADR) files and specialized agent outputs, which are external data sources.
Ingestion points: Processes content from repository-relative ADR markdown files and JSON session data (.adr-session.json).
Boundary markers: Uses structured templates and dedicated lenses (contrarian, user-advocate, meta-process) to isolate and synthesize different perspectives.
Capability inventory: Utilizes Bash for filesystem operations and Task for spawning independent reviewer agents.
Sanitization: Implements strict path validation, explicitly rejecting absolute, traversing, or prose-derived paths, and enforces content hash verification to ensure data integrity between discovery and synthesis.
[COMMAND_EXECUTION]: The skill executes a local script for management tasks.
Evidence: Invokes python3 scripts/adr-query.py to handle path registration and SHA-256 hashing of ADR content.
Safety Context: Execution is limited to local scripts within the repository context, using validated parameters to prevent command injection.
[SAFE]: No obfuscation, unauthorized network operations, or persistence mechanisms were detected. The skill follows best practices for multi-agent orchestration, such as simultaneous dispatch to prevent cross-perspective bias and file-based synthesis for auditability.

adr-consultation