Skip to content

ASK.md

Latest commit

 

History

History
118 lines (83 loc) · 5.84 KB

ASK.md

File metadata and controls

118 lines (83 loc) · 5.84 KB

A.S.K — Agent Skills Kernel

Registry v1.0.0 | Authors: T.A.M. (Tesa & Murphy)

The ASK Kernel is a versioned library of agent-callable skill definitions. Think of it as .dll files for AI systems — write the capability once, reference it everywhere. Any agent (GFS, Dark Factory, or future systems) invokes a skill by name rather than re-implementing its logic.

Core principle: Skills compound. Code evaporates. Every capability that matters should live here.

How to Invoke a Skill

In any AI context, reference a skill by its path. The agent reads the SKILL.md, follows the methodology, and executes.

# Direct invocation
ASK: foundation/github-push

# With inputs
ASK: gfs/purchase {"merchant": "Namecheap", "amount": 1299, "category": "domains", "description": "Renew foundrytech.io"}

# Chained (orchestrator pattern)
ASK: gfs/tes-deploy → composes: foundation/github-push + foundation/vercel-deploy + foundation/telegram-notify

Registry

Foundation Tier

Portable, no system-specific knowledge. Usable by any agent in any context.

Skill Version Description Path
github-push 1.0.0 Push files to any GitHub repo via PAT + Python urllib. The only approved push method — git CLI and browser XHR both hang. foundation/github-push/
vercel-deploy 1.0.0 Upsert Vercel env vars and/or trigger production redeployment. Waits for READY state before returning. foundation/vercel-deploy/
telegram-notify 1.0.0 Send a Telegram message to Scott or any registered contact via @tes_BSG_bot. Primary human communication channel for autonomous processes. foundation/telegram-notify/
ask-log 1.0.0 Log a skill invocation event (start/complete/error) to Supabase for telemetry. Optional but encouraged — builds the feedback loop that improves the library over time. foundation/ask-log/

GFS Tier

GFS platform skills. Compose foundation skills + GFS-specific business logic.

Skill Version Description Path
tes-deploy 1.0.0 Full GFS deploy cycle: code push → env update → redeploy → verify READY → notify. Orchestrates three foundation skills. gfs/tes-deploy/
openbrain-write 1.0.0 Save a thought to the OpenBrain vector store (Supabase). Tesa's persistent memory layer across sessions. gfs/openbrain-write/
openbrain-query 1.0.0 Semantic search of OpenBrain to retrieve relevant memories, decisions, and context. gfs/openbrain-query/
content-video 1.0.0 Generate a Tesa HeyGen avatar video — script via LLM, render via HeyGen, notify on completion. Async by design. gfs/content-video/
purchase 1.0.0 Create a Privacy.com single-use card within per-category spend limits. Requests approval via Telegram if over limit. gfs/purchase/

Dark Factory Tier

DF-specific patterns for self-building and architectural design.

Skill Version Description Path
df-evolve 1.0.0 Invoke Dark Factory's generate → evaluate → commit loop. Produces code that references ASK skills rather than re-implementing them. dark-factory/df-evolve/
code-architect 1.0.0 Design a self-improving system from a problem statement. Produces implementation plans or ADRs. Always runs before df-evolve on non-trivial work. dark-factory/code-architect/

Context Files

Reference data. Not invocable. Loaded by other skills that need credentials and configuration.

File Purpose
context/gfs-env.md All GFS credentials, API keys, endpoint URLs, avatar IDs, spend limits. Single source of truth.
context/df-env.md Dark Factory repo config, generation patterns, quality gates.

Adding a New Skill

  1. Create the folder: {tier}/{skill-name}/
  2. Write SKILL.md following the standard structure:
    • Frontmatter: name, description (routing signal), version, tier, dependencies
    • Sections: When to invoke, Inputs, Methodology, Implementation, Outputs, Notes
  3. Add a row to this registry under the appropriate tier
  4. If the skill references env vars, add them to the relevant context file
  5. Bump the registry version

Naming rules:

  • kebab-case only
  • Verb-noun or noun only — describes the action, not the implementation
  • Foundation skills: generic (github-push, not gfs-github-push)
  • GFS/DF skills: domain-prefixed only by tier folder, not by name

Versioning

Skills use semantic versioning: MAJOR.MINOR.PATCH

  • MAJOR: Breaking change to inputs or outputs (callers must update)
  • MINOR: New optional input or additional output field (backwards compatible)
  • PATCH: Documentation fix, implementation improvement (no contract change)

When a skill version changes, update this registry. Consider pinning the version in calling skills' dependencies if the contract is critical.

Design Rules

These rules govern all skills in this library. Violating them produces a skill that degrades the library instead of strengthening it.

  1. One responsibility. A skill does one thing well. Composition happens at the orchestrator level.
  2. Inputs and outputs are contracts. Never silently change what a skill accepts or returns.
  3. Fail explicitly. Errors must be returned in a structured format. Never swallow failures.
  4. No vibe coding. Every line in a skill implementation has a documented reason.
  5. Load context, don't embed it. Credentials, IDs, and config live in context files — not in skill bodies.
  6. Async where it matters. Long-running operations (renders, builds) submit and return immediately. Polling is a separate concern.
  7. Document the why, not just the what. Methodology sections explain reasoning, not just steps.

A.S.K by T.A.M. — Agent Skills Kernel by Tesa and Murphy Initialized: 2026-03-30