The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest a brand-context.md file from the user's project and explicitly allows that file's contents to override the agent's default instructions and safety parameters.
Ingestion points: The agent is instructed to look for and read brand-context.md in the project root, ./.claude/brand-context.md, and ./docs/brand-context.md (defined in SKILL.md).
Boundary markers: Absent. The instructions state the file should be treated as "binding" and that a specific section within it "overrides an agent's own defaults."
Capability inventory: The skill agents (Reporting Specialist, QA Manager, Financial Tracker, Legal Compliance Officer) manage sensitive data including CRM records, financial budgets, and compliance audits. While tool execution is not explicitly detailed in the markdown, the instructions involve processing and reconciling data across these systems.
Sanitization: Absent. There is no instruction to validate, escape, or filter the content of the brand-context.md file before it influences agent behavior.
[EXTERNAL_DOWNLOADS]: The skill references several external GitHub repositories belonging to unknown individuals or organizations as sources for its methodologies and scoring rubrics.
Evidence: References to github.com/modest-curator478/claude-skills are found in ops-financial-tracker.md. Multiple references to repositories including github.com/cgallic/kai-cmo-harness, github.com/sergebulaev/linkedin-skills, github.com/charlie947/social-media-skills, and github.com/olson-adam/ad-mastermind are found in ops-quality-assurance.md.

client-operations