The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it ingests external, potentially untrusted documentation files to synthesize project intent and requirements.
Ingestion points: Phase 1b in SKILL.md reads files matching patterns such as docs/adr/*.md, docs/prd/*.md, docs/specs/*.md, docs/rfc/*.md, and root-level {ADR,PRD,SPEC,RFC}-*.md.
Boundary markers: The instructions do not specify explicit delimiters or 'ignore' instructions for the interpolated documentation content when passing it to the LLM or subagents.
Capability inventory: The skill uses the Bash tool to create GitHub issues and the Agent tool to spawn subagents for tasks like brainstorm and setup.
Sanitization: The skill implements a 'conflict detection engine' (defined in references/doc-conflict-engine.md) that validates ingested data against 'locked decisions' in existing GitHub issues, providing a layer of verification before state changes occur.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute gh (GitHub CLI) commands. These commands are used for legitimate repository management tasks such as checking authentication status, listing and creating labels, and creating or editing GitHub issues.
[EXTERNAL_DOWNLOADS]: The skill refers the user to https://cli.github.com to download the GitHub CLI if it is not detected on the system. This is a well-known and official source for the tool.

new-project