samples: add skill_reference_layer — a skill collection's reference layer as an OKF bundle

[gamerey43]

What this PR adds

A worked example of an OKF knowledge bundle that demonstrates the package pattern on the largest producer of agent-read markdown in the wild: a skill collection's reference layer. The OKF spec's references/ convention (§6.3) names this pattern in one line; this PR shows it at its most useful scale.

Path convention: the upstream repo places recipes under samples/<thing>/; this sample follows that pattern by including the pre-rendered bundle as samples/skill_reference_layer/bundle/. The recipe in README.md documents how the bundle was produced.

What's in the bundle

21 typed concepts in 6 type groups, each carrying the OKF trust family:

• Skills (6) — entry-point routing + decision, search, connectivity-check, decide, bootstrap, publish concepts
• Probes (4) — direction, adoption, threat, reversal job descriptions
• Models (2) — three-moments, workflow-shapes
• Templates (2) — decision-record-block, dispatch-contract
• Doctrine (6) — read-before-you-write, provenance-is-first-class, errors-name-their-fix, state-has-a-curator, restate-then-investigate, cost-shaped-budget
• Playbooks (1) — cold-start

Plus the bundle's navigational and decision surface:

• index.md — progressive-disclosure listing, grouped by type
• log.md — dated directory history
• decisions.md (type: Decision Record) — six inline decision-record blocks (D-001..D-006)

Conformance

Every concept file (except index.md and log.md per §8/§9 reserved-filename rules) carries parseable YAML frontmatter with non-empty type. The trust family is two-event: a process: automated parse plus a human: review. sources[] resolves inside the bundle.

Why this matters

The OKF v0.2 spec motivates its references/ convention in a single line. Skill collections are where that convention gets its hardest test: agent-written, agent-read, continuously maintained, load-bearing doctrine. This sample shows the cheapest upgrade: keep every path, add frontmatter + an index + a log, and the collection's reference layer becomes a traversable, trust-marked knowledge graph.

Type taxonomy used in the bundle

The bundle uses six self-explanatory type values per OKF §4.1:

• Skill (proposed for v0.3 — covered separately)
• Probe Prompt
• Model
• Template
• Doctrine
• Playbook

index.md groups the bundle by type.

What's deliberately not in the sample

• No client-side knowledge domain content (the methodology is generic on purpose)
• No real-world skill collection vocabulary (the names are fictional; the methodology is plausible but invented)
• No viz.html — knowledge-catalog's own tool generates those on accepted bundles; the contributor flow generates them post-acceptance

Companion proposal

A separate issue argues for type: Skill as a v0.3 conventional concept type — recommended frontmatter, recommended body sections, and trust-family defaults specific to skills. The bundle demonstrates the proposed shape; the proposal argues for adopting it into v0.3.

What reviewers may focus on

• Whether the type taxonomy (six values) is the right one
• Whether the trust-family values (two-event verification) read as honest for a hand-authored bundle
• Whether the index.md groupings are readable for someone landing in the bundle cold
• Whether the deliberately generic methodology body of the bundle reads as a real corpus or as filler

Small edits reshape: the Read Before You Write doctrine is itself part of the bundle and is the cheapest discipline for review.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.