The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill scans untrusted codebase data (source code, schemas, and models) during the Inventory and Distill phases to generate a product map. A malicious codebase could theoretically contain content designed to influence the agent's analysis or cause the generation of misleading map nodes.
Ingestion points: Repository scanning occurs in Phase 2 of SKILL.md and the Inventory stage of references/extraction-protocol.md.
Boundary markers: The protocol enforces structural constraints (D1 Plumbing Filter, D2 Grain Rules) and requires explicit evidence (D4).
Capability inventory: The skill calls MCP tools such as aporia:apply_scan, aporia:upsert_thesis, and aporia:record_notes to transmit data externally.
Sanitization: The workflow mandates a human-in-the-loop review (Phase 5: Elicit) to confirm interpretations and react to the draft map before finalization.
[DYNAMIC_CONTEXT_INJECTION]: The skill utilizes the ! command syntax in Phase 6 of SKILL.md to execute local git commands (git rev-parse --abbrev-ref HEAD, git rev-parse HEAD, and git status --porcelain) for capturing repository metadata. These commands are benign and serve the legitimate purpose of identifying the codebase version being scanned.

aporia-onboard