The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is instructed to access and scan sensitive environment configuration files (.env*) during its 'Reality Scan' and 'Drift Detection' phases to identify project integrations.
Evidence: In SKILL.md under the 'Phase 2: Reality Scan' section, step 2 explicitly includes scanning for .env* patterns. This creates a risk of exposing sensitive credentials or secrets stored in environment files to the model's context.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it ingests various project files and source code which may contain untrusted data, and uses this data to influence document generation.
Ingestion points: Reads BUILDING.md, CLAUDE.md, README.md, package.json, and codebase files including routes, pages, and components via glob patterns.
Boundary markers: The instructions do not specify any markers or delimiters to separate ingested content from system instructions, nor does it warn the agent to ignore instructions embedded in the scanned files.
Capability inventory: The skill has the capability to read local files and write or overwrite documentation files (VISION.md, SPEC.md, BUILDING.md) at the project root.
Sanitization: There is no evidence of sanitization or validation of the ingested content before it is processed by the agent to draft new specifications or updates.

Reconcile