A portable SKILL.md for AI agents that scales planning, tools, testing, safety, and proof to the task's real risk.
Instead of making every request follow a heavyweight ritual, Adaptive Taskcraft uses three rigor lanes and a five-state execution loop:
ALIGN -> FRAME -> ACT -> PROVE -> DELIVER
It integrates and refines lessons from natural-response prompting, planning, TDD, root-cause debugging, CI repair, frontend and Figma implementation, browser testing, CLI and MCP design, external-service integration, threat modeling, and progressive tool disclosure.
Agent failures often come from opposite extremes:
- too little process: assumptions, unsafe writes, unverified claims;
- too much process: verbose plans, context overload, tools loaded without need, and slow trivial work.
Adaptive Taskcraft chooses the smallest workflow that can still produce trustworthy evidence.
Copy this repository's SKILL.md into your agent's skill directory under adaptive-taskcraft/.
Common layouts include:
~/.hermes/skills/adaptive-taskcraft/SKILL.md
~/.agents/skills/adaptive-taskcraft/SKILL.md
<project>/.agents/skills/adaptive-taskcraft/SKILL.md
Exact discovery rules vary by host. Restart or begin a new session after installation if your agent caches skills.
Load the core SKILL.md first. Read references/capability-modules.md only when the task needs a specialized domain module; this keeps progressive disclosure operational rather than merely aspirational.
Load adaptive-taskcraft for substantive answers, software changes, debugging, reviews, integrations, and multi-step execution. The skill itself decides whether the task needs direct action, a guided workflow, or engineered controls.
English, Simplified/Traditional Chinese, and Japanese are supported. The agent matches the user's language and register while preserving identifiers, commands, logs, and exact error text. Multilingual rules live in references/language-adaptation.md.
- natural answers without filler;
- outcome-based plans with acceptance criteria;
- progressive loading of tools and specialized instructions;
- vertical behavior slices and proportional TDD;
- least privilege, consent, rollback, and threat awareness;
- completion claims backed by fresh evidence.
This repository contains instructions, tests for structural invariants, source acknowledgements, and an MIT license. It does not bundle third-party code or provider-specific plugins.
See SOURCES.md for influences and attribution. Domain-specific modules live in references/capability-modules.md and are loaded only when relevant.
python -m pytest tests/test_skill.py -qMIT. Copyright 2026 MoonsvnLyn and FirmamentalSpring.