A Slack-native AI agent framework that integrates with MCP servers, skills, and tools for task automation within Slack workspaces
Ekko defines 5 tools with explicit schemas and descriptions. However, critical gaps limit production readiness: (1) Tool names lack action verbs, 'load_skill', 'read_skill_file', 'render_artifact' are acceptable, but 'example' is generic and non-descriptive. (2) Descriptions are present but inconsistent in depth; 'example' has only 34 chars (below 50-char baseline). (3) Input schemas use Zod with enums (good), but output schemas are not documented, LLMs cannot predict return structure. (4) Error handling is minimal; responses like '{error: "Unknown skill"}' lack recovery guidance. (5) The 'export_pdf' tool has no visible implementation or schema in the provided code. (6) Parameters like 'html' in render_artifact lack format constraints or validation hints. Overall, definitions are functional but lack the rigor expected of production tools.
Echo back the input — a reference tool to show the shape.
Server-side PDF export tool (only available when Vercel Sandbox is available)
Load the full instructions for one of the available skills listed in your system prompt. Call this when a skill is relevant to the request before acting. Available skills: [dynamic list].
Read a bundled resource file from a skill (its references, CSS tokens, component docs, or templates). Call with no path to list the available files for a skill. Use this to follow file references inside a loaded skill before authoring an artifact.
Deliver an authored HTML artifact (deck, document, or social card from the design-system skill) to the Slack thread as ONE self-contained file. Reference skill assets with skill-root-relative paths (e.g. "core/tokens.css", "brands/base/brand.css", "core/deck-runtime.js"); they are inlined and fonts are embedded automatically. The delivered HTML opens in any browser and exports to PDF from the browser print dialog.
export_pdf tool has no visible schema, implementation, or meaningful description ('Server-side PDF export tool (only available when Vercel Sandbox is available)'). Cannot verify input parameters or output structure.
Output schemas are not documented for any tool. LLMs cannot predict return structure (e.g., does load_skill return {name, instructions} or {skill, body, metadata}?). This forces agents to guess and risks parsing errors.
Error responses lack recovery guidance. 'Unknown skill' error does not suggest calling load_skill or list_skills. Agents cannot self-correct without explicit hints.
Tool 'example' is generic and non-descriptive. Name does not convey action; description is only 34 chars (below 50-char baseline). This tool appears to be a reference implementation and should be removed or renamed to 'echo_text'.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | <=2025-11-25 | v2 |
Parameter 'html' in render_artifact lacks format constraints. Description says 'The full HTML document' but does not specify max length, encoding, or validation rules. LLMs may pass invalid or oversized payloads.