This is the single source of truth for "what ships today vs. what is planned." Public
docs (site + README) lead with present-tense capability and link back to the picture captured
here. ADRs and docs/plans/ remain the historical record; this file is the index built on top
of them — do not duplicate their content, point to them.
Maintenance: when a capability lands or a release cuts, update the relevant row here in the
same change. Trust order when sources disagree: current code → tests → examples → ADRs → plans.
Last reconciled: 2026-05-31 (against master @ v0.21.0, issue #280).
Status legend: Shipped (implemented and in use) · Partial (some shipped; gap noted) · Planned (designed, not implemented) · External (depends on an out-of-tree component, e.g. an org's Master Claw policy).
| ADR | Title | Status | Evidence / notes |
|---|---|---|---|
| 001 | cllama Transport | Shipped | cllama/, internal/cllama/ |
| 002 | Runtime Authority | Shipped | docker compose is sole lifecycle writer; Docker SDK read-only |
| 003 | Topology Simplification & HANDLE | Shipped | HANDLE directive; x-claw.handles |
| 004 | Service Surface Skills | Shipped | internal/describe/, SURFACE, mounted skills |
| 006 | INVOKE Scheduling | Shipped | INVOKE directive; claw api schedule |
| 007 | LLM Isolation via Credential Starvation | Shipped | cllama holds keys; agents get bearer tokens |
| 008 | cllama as Standardized Sidecar | Shipped | auto-injected proxy sidecar |
| 009 | Contract Composition & Policy | Shipped | x-claw.include enforce/guide/reference |
| 010 | CLI Surface Simplification | Shipped | four-verb surface; ADR-022 extends |
| 011 | Canonical-By-Default Scaffold | Shipped | claw init, claw agent add |
| 012 | Master Claw & Fleet Governance | Partial | x-claw.master auto-injects claw-api + bearer/feeds wiring (shipped). Autonomous budget/quarantine/recipe decisions are agent behavior + external policy, not enforced infra. Drift scoring not shipped (see 014). |
| 013 | Context Feeds | Shipped | feeds registry; channel-context, channel-awareness |
| 014 | Telemetry Normalization & claw audit |
Partial | claw audit shipped (cmd/claw/audit.go; columns CLAW/REQ/RESP/ERR/INT/TOOLS/TOOL_ERR/TOK_IN/TOK_OUT/COST_USD/MODELS). No drift_score anywhere in cmd/, internal/, cllama/internal/ — drift scoring is External/Planned, not a shipped column. |
| 015 | claw-api Auth & Scoping |
Partial | principals/scopes and write handlers shipped: POST /fleet/restart, /fleet/quarantine, /fleet/budget/set, /fleet/model/restrict, and schedule controls (pause, resume, skip-next, clear-skip-next, fire) route in cmd/claw-api/handler.go; master principals get read/write verbs in internal/clawapi/principal.go. Caveat: budget/model restrict write governance JSON, but reference cllama/runner enforcement of those files is not wired. |
| 016 | Canonical Social Identity & Conformance Spikes | Shipped | sequential-conformance; TestSpikeRollCall |
| 017 | Pod Defaults & Service Self-Description | Shipped | pod defaults; claw.describe; feed registry |
| 018 | Session History & Persistent Memory Surfaces | Shipped | .claw-session-history/, .claw-memory/; claw history |
| 019 | Model Policy Authority & Declared Failover | Shipped | models/models-defaults merge + image MODEL overlay shipped. Declared failover execution is implemented for key/provider exhaustion via internal/cllama/modelpolicy.go, cllama/internal/agentctx/modelpolicy.go, cllama/internal/proxy/modelpolicy.go, and dispatchCandidates in cllama/internal/proxy/handler.go; tests cover 429/cooldown fallback and no fallback on upstream 500. |
| 020 | Compiled Tool Plane (native + mediated) | Shipped | managed tools via claw.describe; cllama mediated execution (v0.6.x schema) |
| 021 | Memory Plane as Compiled Capability | Shipped | pluggable memory services via claw.describe; channel-memory; ambient recall/retain |
| 022 | Infra Image Lifecycle & Four-Verb Surface | Shipped | claw pull/build/up/down; internal/infraimages/release_manifest.go |
| 023 | Explicit cllama Ingress Surface Matrix | Shipped | internal/cllama/ingress.go compiles OpenAI Chat Completions at /v1/chat/completions and Anthropic Messages at /v1/messages. Anthropic-surface providers: anthropic, synthetic, minimax-portal, kimi-coding, cloudflare-ai-gateway, xiaomi; other providers default to OpenAI chat. Runtime dispatch treats /v1/messages as Anthropic and other POST paths as OpenAI. |
| 024 | Runner Base Refresh from Upstream | Shipped | claw pull runner refresh; built-against/image-id/recipe-sha labels |
(ADR-005 does not exist.)
| Capability | Status | Public doc home | Orphaned? |
|---|---|---|---|
Four-verb operator surface (pull/build/up/down) |
Shipped | quickstart, cli, what-is | No |
claw up as compiler |
Shipped | compilation-principles, anatomy | No |
| Clawfile → Dockerfile | Shipped | clawfile | No |
| claw-pod.yml + pod defaults / spread | Shipped | pod-yaml | No |
| cllama governance proxy | Shipped | cllama | No |
| Credential starvation | Shipped | cllama, what-is, index | No |
| Compiled/managed tool mediation (ADR-020) | Shipped | tools | No |
| Memory plane / pluggable recall (ADR-021) | Shipped | memory | No |
| Session history retention (ADR-018) | Shipped | memory, cli | No |
claw audit + telemetry (ADR-014) |
Shipped | cli, cllama | No |
| channel-memory + awareness digests | Shipped | memory, social-topology | No |
| claw-api + Master Claw (ADR-012/015) | Partial | manifesto, what-is, architecture | Policy decisions external; dynamic budget/model enforcement partial |
| Social topology / HANDLE (ADR-003/016) | Shipped | social-topology | No |
| Persona materialization | Shipped | anatomy | No |
| 7 runner drivers | Shipped | drivers | No |
| Context feeds (ADR-013) | Shipped | surfaces-and-skills, cllama, social-topology | No |
TRACK / recipe / bake |
Planned | what-is roadmap, architecture | Correctly future, label it clearly |
| Drift scoring | External/Planned | what-is, cllama, cli, README, manifesto, architecture | No built-in score; docs mark it external/planned |
Confirmed against code. Historical changelog entries are excluded (incident record — leave
alone): the downstream-deployment references in older site/changelog.md entries (e.g.
:123, :187) stay untouched.
Named-deployment leaks (scrub → generic):
site/index.md:79-88— fixed by Claude in #280 laneREADME.md:213-222— fixed by Claude in #280 lanesite/guide/social-topology.md:71-77— fixed by Codex in #280 lanesite/guide/pod-yaml.md:24-33— fixed by Codex in #280 lane
Drift-score / fabricated-output over-claims (→ describe telemetry/audit as shipped; drift as external/planned):
site/guide/what-is-clawdapus.md— fixed by Claude in #280 laneREADME.md:559-575— fixed by Claude; Codex corrected the audit totals examplesite/guide/cllama.md:123,:359— fixed by Codex in #280 lanesite/guide/cli.md:210,:491— fixed by Codex in #280 laneMANIFESTO.md:152,:161— fixed by Claude in #280 lane
Stale phase framing: every "Phase N" mention → reframe present-tense or move to roadmap.
Tier C = correctness pass (scrub names/fabricated output/stale phases, verify CLI snippets). Tier D = deep rewrite (mental model stale/weak).
| Page | Tier | Owner | Status | Notes |
|---|---|---|---|---|
| docs/PROJECT_STATE.md | — | Claude | done | this file (living) |
| site/guide/architecture.md (new) | D | Claude | done | capabilities ↔ ADRs; commit 71227a6 |
| site/index.md | D | Claude | done | leak + Master Claw card; commit ada159b |
| site/guide/what-is-clawdapus.md | D | Claude | done | real audit output, drift honest, de-phased; ada159b |
| site/manifesto.md | D | Claude | done | recipe→roadmap, drift clarifier, link fix; 45cb05c |
| MANIFESTO.md (root) | D | Claude | done | same as site manifesto; 45cb05c |
| README.md | D | Claude | done | leak + audit + retired v0.3.2/phase table; b4080d6 |
| site/.vitepress/config.mts | C | Claude | done | architecture in Introduction sidebar; 71227a6 |
| site/guide/cllama.md | D | Codex | done | spec-vs-reference split; context mount; telemetry; no DRIFT/private-SSH claims |
| site/guide/cli.md | D | Codex | done | real audit columns/totals; no invalid --last; history/discover/skill/update refs; memory forget requires --agent; planned commands marked unregistered |
| site/guide/social-topology.md | D | Codex | done | leak scrubbed; platform matrix refreshed |
| site/guide/pod-yaml.md | C | Codex | done | leak scrubbed; service/default fields refreshed; channel-context tuning clarified |
| site/guide/anatomy.md | C | Codex | done | proxy wording, NanoClaw runtime, durable state split |
| site/guide/clawfile.md | C | Codex | done | TRACK/recipe/bake marked planned; PRIVILEGE syntax fixed; matrix refreshed |
| site/guide/tools.md | C | Codex | done | readOnly/audit claim fixed; current stdio image tag |
| site/guide/memory.md | C | Codex | done | claw history export, required --agent, channel-memory/awareness coverage |
| site/guide/surfaces-and-skills.md | C | Codex | done | egress surface, planned skillmap, RailsTrail as planned adapter pattern |
| site/guide/compilation-principles.md | C | Codex | done | actual projections/manifests; RailsTrail framed as planned pattern |
| site/guide/drivers.md | C | Codex | done | 7-driver matrix refreshed against internal/driver/* |
Out-of-scope leak found (needs its own decision): examples/trading-desk/ itself uses
named-deployment service names and agent files — a public-artifact leak larger than the doc
pages. Renaming ripples into
agent paths, env vars, and possibly spike fixtures, so it is not included in this docs PR;
flag to the maintainer as a follow-up issue.
Release guardrails: no changes to release_manifest.go pins, changelog Latest badge, nav
version dropdown, or cllama submodule pointer. A ## Unreleased changelog note is allowed.