A methodology plugin for AI coding agents. One connected topology from problem framing through shipped change to closed loop.
Groundwork owns the protocols, skills, artifact schemas, and topology. It registers with runa through a single manifest file and has no runtime, no CLI, no installer.
Reliable engineering has a discipline: constraints get verified before design, behavior gets defined before code, completion gets proven before it's claimed, and changes land with merge, cleanup, and issue resolution. This discipline is what makes outcomes repeatable. Groundwork encodes it as a connected set of protocols and skills, giving AI agents the structure to produce verified, shipped work without the need for human course-correction — not through restrictions, but through a topology that makes the right sequence the natural one.
Five stages form the forward flow. Cross-cutting skills fire at any stage, creating branches and feedback loops. The behavior contract from stage 2 is the central integration thread — it flows through design, execution, verification, and landing.
| Stage | Protocols | Produces |
|---|---|---|
| 1. Survey and select work | survey, decompose, begin |
Assessment, executable issues, session scope |
| 2. Define behavior | specify |
behavior-contract — threads through all subsequent stages |
| 3. Design | plan |
Decision-complete design |
| 4. Execute and verify | test, document, verify, propose |
Test evidence, documentation record, completion evidence, open PR |
| 5. Land | land |
Completion record — merge, cleanup, coverage status, issue closure |
Enter where the work needs you. A bug with an existing issue enters at Execute and verify. A new capability enters at Survey and select work. The constraint is sequence — you can't land before executing — not completeness.
These fire at any stage when their trigger condition appears, not at a fixed position:
orientprovides methodology orientation and carries the documentation discipline the agent should keep active at all timesreckonre-fires on any new generative act (design, spec, architecture) — not step-one-onceresearchfires when a decision needs evidence outside the codebasedebugfires on failures; hands off totest(fix),reckon(3-fix escalation), orresolve(environmental cause)resolvefires on operational friction; resolves structurally or files an issue viadecomposecontractcarries the behavior contract through implementation and verification so execution does not drift from specified behavior
Handoff contracts between protocols and skills are defined in topology-contract.md. The issue persistence model is defined in issue-model.md.
| Protocol | Trigger |
|---|---|
survey |
when the work is under-specified and needs a grounded assessment before selection or decomposition |
decompose |
creating, refining, or closing executable issues from an assessment or work definition |
begin |
initiating a work session: selecting work, preparing workspace, declaring direction |
specify |
when defining or refining behavior expectations |
plan |
implementation needs design convergence — multiple approaches, unclear scope, or cross-cutting changes |
test |
when implementing any feature or bugfix — RED → GREEN → REFACTOR |
document |
after code changes that may cause drift, before verification, or when docs need classification |
verify |
before claiming work is complete, fixed, or passing — evidence first |
propose |
packaging changes for review: propose, submit pr, create pr, open pr, send for review |
land |
merge-and-close completion events: land, merge and close, ship it |
| Skill | Trigger |
|---|---|
orient |
session start, task initiation, or any moment requiring methodology orientation |
reckon |
before creating designs/specs/architectures/processes |
research |
when reliable external evidence is needed for decisions |
debug |
when a test fails or behavior is unexpected, before proposing any fix |
resolve |
operational friction — missing tools, broken configs, stale conventions, undocumented requirements |
contract |
carrying a behavior contract through implementation and verification without drift |
| File | Purpose |
|---|---|
groundwork.toml |
Canonical manifest — all artifact types and protocol declarations with interface edges |
docs/architecture/topology-contract.md |
Formal handoff contracts and anti-divergence rules |
docs/architecture/issue-model.md |
Issue state model, dependency graph format, graph maintenance |
schemas/ |
JSON Schema contracts for artifact types |
protocols/ |
Protocol definitions — runa-managed stages declared in the manifest |
skills/ |
Skill definitions — agent-managed cognitive tools, each a SKILL.md with YAML frontmatter |
groundwork.toml # Methodology manifest — artifact types, protocol declarations
schemas/ # JSON Schema contracts for artifact types
protocols/ # Runa-managed topology stages (SKILL.md + references)
survey/ # territory assessment
decompose/ # issue lifecycle
begin/ # work initiation
specify/ # behavior contract definition
plan/ # design convergence
test/ # RED-GREEN-REFACTOR execution
document/ # documentation review/update
verify/ # completion gate
propose/ # commit, push, PR creation
land/ # closeout workflow
skills/ # Agent-managed cognitive tools (SKILL.md + references)
orient/ # methodology orientation
reckon/ # first-principles reckoning
research/ # external evidence gathering
debug/ # root-cause investigation
resolve/ # friction resolution
contract/ # behavior-contract discipline
docs/
architecture/ # Topology contract, issue model, ADRs
tests/
fixtures/artifacts/ # Valid/invalid artifact examples for schema testing