The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines an attack surface for indirect prompt injection as it ingests untrusted external data such as 'market signals' and 'stakeholder intent'. However, this risk is mitigated by the design of the skill.
Ingestion points: External vision, market signals, and stakeholder intent are processed via prompt instructions in SKILL.md.
Boundary markers: The 'Business Gate 0' section explicitly mandates human approval ('Do not proceed to downstream skills until Human approves') before the strategy is finalized or acted upon.
Capability inventory: The skill's capabilities are restricted to writing static Markdown artifacts and YAML manifests to the .agile-v/ local directory. It lacks network access or shell execution capabilities.
Sanitization: Human curation and validation are required for all artifacts and assumptions before they are committed to the pipeline.
[SAFE]: The skill limits its file operations to project-specific paths (.agile-v/business/ and .agile-v/aibom/) for documentation and tracking purposes, avoiding sensitive system directories.
[SAFE]: All referenced tools, conventions, and metadata (e.g., agile-v-core, agile-v-aibom) are legitimate resources associated with the skill's author, agile-v.

venture-strategist