Sync Boatstack from Intelligence Flow Labs @ 5f4dda1585d8

[operator-stack-publisher]

What this sync releases

Boatstack helps repositories that cannot yet produce evidence

When a feature needs visual evidence but the repository has no command to produce
it, Boatstack now guides you to set one up instead of silently leaving a gap. A
new provision-capability operation detects the repository's frontend stack
(framework, package manager, and whether Playwright, Cypress, or Storybook are
present) and returns a context-aware guide: the framework-agnostic contract the
in-repository capture harness must satisfy, plus stack-tailored steps to build
it. Boatstack ships the contract and the guide, never the harness itself.

capability-register then records the repository-owned command in one step,
keeping the canonical .boatstack-project.json source and every generated file
in sync (or round-tripping the generated configuration alone when there is no
source). Once registered, provisioning reports the capability as available and
capture-evidence can run.

Planning is aware of this too: when a visual scenario is relevant but no capture
command resolves, auto-plan surfaces a material provisioning decision with tiered
paths — provision now as its own delivery slice, bundle the harness into the
feature slice, or record the gap and defer. It remains a surfaced choice;
Boatstack never imposes a frontend framework.

Boatstack can now capture PR visual evidence for you

A new capture-evidence operation turns a repository's declared visual scenarios
into trusted PR evidence without manual screenshotting. It resolves the
repository-owned capability command, reads the scenarios recorded in the feature
plan, and runs each one as a supervised, fingerprinted operation — retrying a
flaky capture within a bounded budget and reusing a successful capture on the
same commit instead of re-running it.

Each screenshot the harness produces is conformance-checked before it is
ingested: the manifest is stamped to the current head commit and product diff, so
pr-context trusts it as PASS only while it still matches the change under
review. A harness that reports success but produces a non-conformant artifact
fails closed — capture never records evidence it cannot stand behind.

Boatstack ships the capture contract, not the harness. The repository owns a
visual (or screenshot/e2e) command that renders one PNG per scenario, and
Boatstack invokes it through a stable environment-variable contract
(BOATSTACK_CAPTURE_SCENARIO_ID, _ENTRY, _STATE, _VIEWPORT, _OUTPUT).

Evidence capabilities are now a shared, extensible substrate

Boatstack's PR evidence detection is no longer hard-wired to a single evidence
type. A generic evidence-capability registry now backs the "can this repository
produce the evidence?" cut, and visual evidence is its first tenant. Nothing
changes in how visual evidence behaves today — the same repository commands,
statuses, and fingerprints — but the shared spine means future evidence types
(and the upcoming capture and provisioning flows) plug in through one registry
instead of a bespoke path each time.

Publishing toolkit moved to a standalone package

The labkit toolkit that projects Boatstack into this repository moved from a
submodule of the monorepo's Python package to a standalone top-level labkit/
project, invoked as python -m labkit. This only changes where the publishing
tooling lives and how the sync workflow installs it; the projected files are
byte-for-byte identical (asserted by the golden reproduction test), with no
effect on Boatstack's runtime, CLI, skill, or public contract.

Projection provenance

Generated from operatorstack/intelligence-flow@5f4dda1585d8032019dd0ed2c3b8ea264abfcf4d.
Review provenance, tests, examples, and context-cost changes before merging.

[bigboateng]

Closing: superseded stale sync. Intermediate snapshot from the old workflow; conflicts with main after the labkit control-plane migration (#83) and upstream has advanced past this commit. A fresh engine-driven sync for the current tip replaces it.