Capability classification (for agents and reviewers)

Use this vocabulary when describing features in PRs, docs/implementation-log.md, and code comments. Do not label something “complete” without evidence from the verification column.

States

State	Meaning	Evidence bar (minimum)
complete	Matches the documented contract; safe for default CI and documented workflows.	Code path exists; default tests (just ci) or clearly scoped tests cover it; docs/commands match behavior.
partial	Shipped and safe, but missing depth, optional deps, or edge cases—call out what is missing.	Tests or manual steps document the gap; README/AGENTS do not imply full coverage.
stub	Intentional non-implementation: raises NotImplementedError, or exists only to reserve an API.	Raised error or obvious no-op; referenced from architecture or this file; not described as production-ready elsewhere.
documented-only	Described in docs or comments but no code path, or only scaffolding.	No contradictory “complete” claims; prefer upgrading to stub with an explicit error.
broken	Known incorrect behavior or failing tests; not acceptable on main without a ticket/fix plan.	Failing test, issue link, or explicit “do not use” in docs.

Cross-check: “complete” vs hype

Before writing complete:

1. Code — Grep for the public entrypoint (cli.py, create_*, factory).
2. Tests — tests/ (and tests/integration/ only if the feature is integration-only).
3. Commands — alwm … --help or just … matches README / AGENTS.
4. Optional deps — If the feature needs uv pip install -e '.[browser]' (or similar) for optional behavior, say partial or complete (optional extra) and name the extra.

Repository examples (current; re-verify after changes)

Capability	Suggested label	Why (evidence)
alwm validate kinds	complete	artifacts.py + tests/test_domain.py / smoke
Committed examples/ / fixtures/ contract drift guard	complete	just validate-artifacts (tests/test_schema_drift_contracts.py); inventory docs/audits/schema-drift-contracts-inventory.md
MockBrowserRunner / FileBrowserRunner	complete	tests/test_browser.py; no network
PlaywrightBrowserRunner	partial (optional)	Requires [browser] extra + playwright install; tests/integration/ gated by ALWM_PLAYWRIGHT_SMOKE=1
MCPBrowserRunner	partial	Fixture path or optional local stdio MCP when ALWM_MCP_BROWSER_COMMAND is set (browser/mcp_runner.py, browser/mcp_stdio.py); requires mcp (dev extra); alwm browser run-mcp; default tests cover the shipped stdio fixture server (tests/test_mcp_stdio.py); remote / IDE-hosted MCP not in scope (docs/roadmap/v0.2.0.md, docs/architecture/browser.md)
browser_realism.v1 rubric	partial	examples/dataset/rubrics/browser_realism.v1.json — scores grounding in committed BrowserEvidence JSON; does not assert a live browser unless evidence came from Playwright (optional [browser])
alwm benchmark run (offline)	complete	tests/test_benchmark.py, tests/test_benchmark_browser.py; browser_mock variants run browser phase + write browser_evidence.json
alwm benchmark campaign run (incl. --dry-run)	complete	tests/test_benchmark_campaign.py; manifests registered in artifacts.py; workflow docs/workflows/benchmark-campaigns.md, walkthrough docs/workflows/campaign-walkthrough.md
Benchmark + Playwright	partial (opt-in)	Variant browser.runner: playwright requires ALWM_BENCHMARK_PLAYWRIGHT=1 + [browser] extra; not in default CI
alwm benchmark probe	complete (live)	tests/test_live_probe.py; integration opt-in
Prompt registry CLI	complete	tests/test_prompt_registry.py. Registry content grows over time—that is data, not stubbing.

When to file drift

If README, AGENTS.md, docs/workflows/verification.md, and docs/architecture/*.md disagree about a verification command or dependency, treat it as drift: fix docs or code in the same PR, or open a short note in docs/implementation-log.md.