Portable protocols for agent memory and coordination.
agent-wiki is an agent-first, machine-readable knowledge base for memory, workspace, skill lifecycle, deliberation, and coordination. Start with one problem and the smallest useful binding.
Smallest reliable reading path
01 · Route
protocols.yaml
02 · Bind
one minimum level
03 · Calibrate
one matching use case, when available
- 7 protocols
- 5 guides
- 12 documented use cases
- CC BY 4.0
Start with the constraint you actually have.
The Adoption Guide is the source for this routing table. Selection does not authorize environment changes, and adding a protocol is unnecessary when a simpler action already solves the problem.
- Useful context must survive sessionsMemory · memory:L0
- A user wants a failed run or prior conclusion to stop influencing future workMemory · memory:L0
- A corpus needs weak lateral explorationAuto-Walk · auto-walk:L0
- Imported or relationship-formed procedures need governed adoption and evolutionSkill Lifecycle · skill:L0
- The wrong information is active during a taskActive Workspace · active-workspace:L0
- An Agent needs a short self-guidance cueInner Speech · inner-speech:L0
- A decision needs inspectable competing viewsCouncil · council:L0
- One human wants one interface over many executorsSteward · steward:S1
- Several chosen protocols must cooperateComposition guide · composition:C1
- Several participants need governed shared standing over existing artifactsGoverned Shared Memory profile · No level; design-only guide
- Governed artifacts or a binding must be backed up, restored, or movedGoverned Artifact Portability and Recovery guide · No level; design-only guide
- Governed artifacts must remain available across independently usable bindings while updates continueGoverned Artifact Replication and Exchange guide · No level; design-only guide
Choose the smallest sufficient binding.
Levels are cumulative within one protocol and are not comparable across protocols. Versions below 1.0.0 may change incompatibly.
| Protocol | Version | Maturity | Minimum | Usecase conformance |
|---|---|---|---|---|
| Agent-first MemoryUseful context must survive sessions | 0.2.0 | practiced | memory:L0 | mapped 3 |
| Agent-first Auto-WalkA corpus needs weak lateral exploration | 0.1.0 | practiced | auto-walk:L0 | proposed 3 · partially-verified 1 |
| Agent-first Skill LifecycleImported or relationship-formed procedures need governed adoption and evolution | 0.1.0 | practiced | skill:L0 | mapped 1 |
| Agent-first Active WorkspaceThe wrong information is active during a task | 0.2.0 | design-only | active-workspace:L0 | 0 documented |
| Agent-first Inner SpeechAn Agent needs a short self-guidance cue | 0.1.0 | design-only | inner-speech:L0 | 0 documented |
| Agent-first CouncilA decision needs inspectable competing views | 0.1.0 | practiced | council:L0 | mapped 3 |
| Agent-first StewardOne human wants one interface over many executors | 0.1.0 | practiced | steward:S1 | partially-verified 1 |
Three axes. No borrowed certainty.
Maturity describes the protocol, evidence describes an observation boundary, and conformance describes checks against a named protocol version. One does not silently upgrade another.
Maturity
How far a protocol has moved from bounded design into repeated operational feedback.
design-only · practiced · field-tested
design-only 2 · practiced 5 · field-tested 0
Evidence
What kind of source inspection, run report, reproduction, or field operation supports a use case.
design-example · source-inspected · run-reported · reproduced · field-tested
design-example 3 · source-inspected 1 · run-reported 2 · reproduced 0 · field-tested 6
Conformance
How directly the named requirements were mapped, inspected, or executed, with declared gaps preserved.
proposed · mapped · partially-verified · verified
proposed 3 · mapped 7 · partially-verified 2 · verified 0
12 manifest-listed use cases · 0 verified conformance claims. Read each evidence scope before reusing a conclusion.
One protocol. One bounded attempt.
Partial and negative results are useful. Record the route, runtime, observed result, remaining gaps, and friction without upgrading the claim.
Read the manifest
Read protocols.yaml for protocol ids, versions, maturity, dependencies, artifacts, and usecase evidence.
Select by problem
Select one protocol from the problem table below.
Read the complete contract
Read that protocol's metadata, Purpose, Non-goals, core artifact/schema, minimum binding, implementation levels, validation checklist, and Final rule.
Calibrate with one use case
Read one matching use case, when one exists, and calibrate claims from its Evidence and Validation scope fields.
Compose only when required
Read Composable Agent Cognition Protocols only when two or more protocols must exchange artifacts.