The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple sources including user-provided briefs, repository documentation (e.g., AGENTS.md), and tracker content (e.g., PR bodies, issue descriptions).
Ingestion points: {brief} argument, .ai/agentic.config.json, repository agent docs, tracker content, and external knowledge sources provided via knowledge.sources.
Boundary markers: The skill explicitly defines an 'Untrusted content boundary' in references/agentic-setup.md, instructing the agent to treat external inputs as data and never as instructions.
Capability inventory: The skill performs file reads, writes App Spec documents to ${SPECS_DIR}, dispatches subagents for review, and invokes other skills such as om-spec-writing and om-gap-analysis.
Sanitization: Instructions in references/agentic-setup.md require validation of externally-sourced values using a specific regex (^[A-Za-z0-9._/-]+$) before any shell or path interpolation.
[COMMAND_EXECUTION]: The skill utilizes local shell commands to parse its configuration files during the setup phase.
Evidence: Found in references/agentic-setup.md, where a bash script snippet uses jq to extract configuration values like SPECS_DIR and PLATFORM_REPO from a local JSON file.
[DYNAMIC_EXECUTION]: The skill dispatches fresh-context subagents to perform critical reviews and architectural checkpoints.
Evidence: Described in SKILL.md (Challenger gate and Architect checkpoint) and supported by the instructions in references/challenger-prompt.md. This constitutes a controlled delegation of auditing tasks to secondary agents.

om-app-spec-writing