Adaptive Conformance Specification (ACS) Skill
A foundational behavioral protocol for agents working within unfamiliar codebases or tool surfaces. ACS defines how an agent must observe, infer, and conform to a system before generating any artifact within it.
ACS produces no artifact of its own. It produces disciplined behavior that shapes whatever the invoking skill produces.
Two downstream skills motivated this specification:
Adaptive Tool Discovery (ATD) maps the capabilities of an external tool set: what tools exist, what each does, how to call them correctly, and what their failure modes are. The output is a tool capability map the agent can act on.
Adaptive Onboarding (AO) builds the context needed to work as a member of a specific team or project — conventions, ownership, local definition of done, and norms not written down anywhere. The output is team-member-level working context.
Both skills require the same foundational discipline: observe before acting, infer from evidence, conform to local patterns, stop rather than guess. ACS extracts that shared layer so it is defined once, maintained once, and invoked consistently by both.
- The evidentiary standard an agent must meet before generating output
- A six-step workflow: Discover → Interpret → Select → Execute → Validate → Conclude
- Principles for conforming to local patterns, including when outside conventions are permitted
- Error detection and structured reporting
- Stopping conditions for insufficient evidence
- An invocation protocol for dependent skills
- Which concerns to extract and where to find them (defined by the invoking skill)
- Tool capability mapping → see adaptive-tool-discovery
- Team and project onboarding context → see adaptive-onboarding
- Post-conformance verification of agent understanding
ACS is not invoked directly by users. It is invoked by skills that need its observation and conformance layer.
Skills that depend on ACS include this in their preamble:
This skill operates under the Adaptive Conformance Specification (ACS). Apply ACS discovery and conformance steps before executing any domain-specific actions below.
If you are building a skill that requires an agent to work within an unfamiliar
system, ACS is the correct foundation.
Read SKILL.md for the full specification.
Clone the repo and add your own private scenarios under evaluation/local/.
That directory is gitignored and will never be committed or pushed.
Your proprietary test cases stay local while you pull upstream updates freely.
git clone https://github.com/adaptive-interfaces/adaptive-conformance-specification
cd adaptive-conformance-specification
mkdir -p evaluation/local/my-scenarioSee evaluation/scenarios/ for examples of how
scenarios are structured.
adaptive-conformance-specification/
SKILL.md the specification itself
MANIFEST.toml repository intent, scope, and role
DECISIONS.md design history and rationale
LICENSE MIT
evaluation/
rubric.md grading criteria for conformance quality
scenarios/ public test cases
local/ gitignored, private test cases
Format Markdown files with Prettier extension. Then run:
npx markdownlint-cli2 "**/*.md"
uvx skillcheck SKILL.md --min-desc-score 75MIT © 2026 Adaptive Interfaces