The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses dynamic context injection (the ! command syntax) in SKILL.md to execute git rev-parse --show-toplevel. This is used legitimately for repository discovery and identifying the project root at load time.
[COMMAND_EXECUTION]: The execution logic involves dispatching specialized sub-agents (e.g., ce-slack-researcher, extraction-tier scouts) whose instructions are dynamically loaded from local markdown files in the references/ directory.
[EXTERNAL_DOWNLOADS]: The HTML rendering reference facilitates fetching webfonts from well-known services, specifically Google Fonts and Bunny Fonts, to maintain document typography when viewed externally.
[DATA_EXFILTRATION]: The skill includes an optional handoff to the ce-proof utility, which enables users to publish the generated requirements document to an external service (Every's Proof) for sharing and collaboration.
[PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it processes untrusted data from repository files and Slack conversations.
Ingestion points: Phase 1.1 performs automated searches and reads across the local repository; the ce-slack-researcher persona reads public and private Slack threads.
Boundary markers: The persona instructions include specific 'Untrusted Input Handling' rules that direct the agent to treat message content as data for extraction rather than as authoritative instructions.
Capability inventory: File system reads, sub-agent spawning, and network operations (Slack API access and Proof document uploads).
Sanitization: The skill employs interpretive principles to filter Slack noise and focus on factual decisions or constraints.

ce-brainstorm