The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests untrusted data from intake files and external web search results to inform project constraints and terminology.
Ingestion points: Processes content from 01-intake.json, 02-context.json, and dynamic web search results when search_mode is set to web or mixed.
Boundary markers: The skill employs rigorous boundary markers including a mandatory JSON schema (02b-evidence.schema.json), explicit used_for allowed-lists, and a 'Hard boundary' section in SKILL.md that strictly limits the influence of found evidence on the agent's scope.
Capability inventory: Capabilities include network operations for web searching, filesystem writes to the internal/stages/ directory, and execution of the local validation script i2r.py.
Sanitization: Implements sanitization via the i2r-citation-gate which executes i2r.py evidence.validate to verify all sources and ensure the generated output conforms to the required safety and structural standards.
[EXTERNAL_DOWNLOADS]: The skill performs external network requests to search for documentation and evidence. These operations are restricted by the search_mode configuration and are primary to the skill's intended purpose of gathering external domain framing and regulatory constraints.

i2r-search-mode