The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests project-specific data including technical specifications, user stories, and evaluation examples from the docs/specs/ and .claude/conclave/eval-examples/ directories to guide the behavior of the spawned agent team.
Ingestion points: SKILL.md reads content from specification files and user-provided evaluation examples to calibrate the judgment of the Quality Skeptic and other agents.
Boundary markers: External content is interpolated into agent prompts using markdown headers and fenced code blocks, though it lacks explicit instructions to ignore instructions embedded within that data.
Capability inventory: The specialized agents have the ability to write project source code and execute testing tools (e.g., Playwright) as part of their primary functions.
Sanitization: No explicit filtering or sanitization of the markdown content is performed prior to interpolation into the teammate prompts.
[COMMAND_EXECUTION]: The guardrails.md file defines a comprehensive set of static analysis, linting, and testing tools (such as Larastan, ESLint, Ruff, and Semgrep) that the team is expected to run. These commands are standard engineering tools and their use is governed by the framework's quality standards.

build-product