odibi-anchor 0.1.0__tar.gz

odibi_anchor-0.1.0/.assistant/README.md

@@ -0,0 +1,23 @@
+# Native skill distribution
+
+Every `SKILL.md` must contain, in frontmatter and body:
+
+- Frontmatter: `name:` (matches directory) and `description:` (one line, used by `anchor("skills")`).
+- `## When to load` — a specific trigger (user request or symptom).
+- `## When NOT to load` — at least one explicit exclusion.
+- `## Enforcement` — one of: `System hook` (a gate enforces this),
+  `Self-enforced` (discipline), or `Advisory`.
+- Cross-references to related skills via `[[skill-name]]` or `skills/<name>/SKILL.md`.
+
+The `skills/` directory contains exactly the native skill packages. Deterministic
+Odibi Anchor documentation, universal development guidance, and lifecycle detail
+live under `references/`; they are global distribution resources rather than skills.
+
+`references/registry.json` additionally indexes the Engineering Reference Library.
+Use `anchor("references")` for compact discovery, `anchor("references", "match", "query")`
+for deterministic routing, and `anchor("references", "load", "id")` to load one complete
+authored reference. Use `anchor("references", "search", "query", reference_id="id")` to search
+licensed offline snapshots and `anchor("references", "load-section", "sec-v1-...")` to load one
+exact bounded source section. References are advisory technical knowledge, not skills or gate obligations.
+Registry entries include pinned licensed offline source snapshots; external URLs are provenance and
+refresh locations, not a runtime dependency.

odibi_anchor-0.1.0/.assistant/agent_bootstrap.py

@@ -0,0 +1,125 @@
+"""Resolve one Odibi Anchor checkout and run its authoritative bootstrap."""
+
+from __future__ import annotations
+
+if "__file__" not in globals():
+    raise RuntimeError(
+        "Odibi Anchor bootstrap requires __file__; use "
+        "runpy.run_path(<exact .assistant/agent_bootstrap.py>) in the persistent Python process"
+    )
+
+import os
+import runpy
+from collections.abc import Mapping
+from pathlib import Path
+
+
+def _validated_checkout(raw: object, *, source: str) -> Path:
+    if not isinstance(raw, (str, os.PathLike)) or isinstance(raw, bytes):
+        raise RuntimeError(f"{source} must be an absolute, non-tilde path")
+    value = os.fspath(raw)
+    if not isinstance(value, str) or not value or value.startswith("~") or not Path(value).is_absolute():
+        raise RuntimeError(f"{source} must be an absolute, non-tilde path")
+    checkout = Path(value).resolve(strict=False)
+    markers = (
+        checkout / "agent_bootstrap.py",
+        checkout / "src" / "odibi_anchor" / "bootstrap.py",
+    )
+    if not checkout.is_dir() or not all(marker.is_file() for marker in markers):
+        raise RuntimeError(
+            f"{source} is not a Odibi Anchor source checkout: {checkout}; "
+            "expected agent_bootstrap.py and src/odibi_anchor/bootstrap.py"
+        )
+    return checkout.resolve(strict=True)
+
+
+try:
+    _launcher = Path(__file__).resolve(strict=True)
+except (OSError, TypeError) as exc:
+    raise RuntimeError("Odibi Anchor launcher __file__ is invalid") from exc
+
+if "ANCHOR_SOURCE_CHECKOUT" in globals():
+    _checkout = _validated_checkout(
+        globals()["ANCHOR_SOURCE_CHECKOUT"], source="ANCHOR_SOURCE_CHECKOUT init global"
+    )
+elif "ANCHOR_SOURCE_CHECKOUT" in os.environ:
+    _checkout = _validated_checkout(
+        os.environ["ANCHOR_SOURCE_CHECKOUT"], source="ANCHOR_SOURCE_CHECKOUT environment variable"
+    )
+else:
+    _assistant_parent = _launcher.parent.parent
+    _source_candidate = _assistant_parent
+    _copied_candidate = _assistant_parent / "odibi_anchor"
+    try:
+        _checkout = _validated_checkout(_source_candidate, source="source-tree topology")
+    except RuntimeError:
+        try:
+            _checkout = _validated_checkout(_copied_candidate, source="copied-tree topology")
+        except RuntimeError:
+            _checkout = None
+
+if _checkout is None:
+    from odibi_anchor import launch
+
+    _home = os.environ.get("ANCHOR_HOME")
+    if not _home:
+        raise RuntimeError(
+            "Installed Odibi Anchor startup requires one explicit durable ANCHOR_HOME"
+        )
+    _target = os.environ.get("ANCHOR_PROJECT_ROOT") or str(_launcher.parent.parent)
+    anchor = launch(
+        anchor_home=_home,
+        project_id=os.environ.get("ANCHOR_PROJECT_ID"),
+        project_root=_target,
+        output_format="dict",
+    )
+    ROOT = str(Path(_target).resolve())
+    MANIFEST = getattr(anchor, "manifest", None)
+    ORIENTATION = anchor("orient", output_format="dict")
+    if not isinstance(ORIENTATION, Mapping) or ORIENTATION.get("kind") != "orientation":
+        raise RuntimeError("Odibi Anchor orientation returned an invalid structured result")
+    _status = ORIENTATION.get("status")
+    _runtime = _status.get("runtime") if isinstance(_status, Mapping) else None
+    _binding = _runtime.get("route_binding") if isinstance(_runtime, Mapping) else None
+    _project_id = _binding.get("project_id") if isinstance(_binding, Mapping) else None
+    BOOTSTRAP = {
+        "success": True,
+        "kind": "agent_bootstrap",
+        "runtime": "installed_distribution",
+        "repository": ROOT,
+        "state_home": str(Path(_home).resolve()),
+        "project_id": _project_id,
+        "target_root": ROOT,
+    }
+else:
+    _delegate_globals = {}
+    if "ANCHOR_REPOSITORY_PROVIDER" in globals():
+        _delegate_globals["ANCHOR_REPOSITORY_PROVIDER"] = globals()["ANCHOR_REPOSITORY_PROVIDER"]
+    _namespace = runpy.run_path(
+        str(_checkout / "agent_bootstrap.py"),
+        init_globals=_delegate_globals,
+        run_name="__odibi_anchor_checkout_bootstrap__",
+    )
+
+    _required = ("anchor", "ROOT", "MANIFEST", "ORIENTATION", "BOOTSTRAP")
+    _missing = tuple(name for name in _required if name not in _namespace)
+    if _missing:
+        raise RuntimeError(f"Odibi Anchor bootstrap omitted required values: {', '.join(_missing)}")
+    _bootstrap = _namespace["BOOTSTRAP"]
+    if not isinstance(_bootstrap, Mapping) or _bootstrap.get("success") is not True:
+        raise RuntimeError("Odibi Anchor bootstrap did not report success")
+    try:
+        _reported_repository = Path(_bootstrap["repository"]).resolve(strict=True)
+    except (KeyError, OSError, TypeError, ValueError) as exc:
+        raise RuntimeError("Odibi Anchor bootstrap reported an invalid repository") from exc
+    if _reported_repository != _checkout:
+        raise RuntimeError(
+            "Odibi Anchor bootstrap repository does not match the selected checkout: "
+            f"{_reported_repository} != {_checkout}"
+        )
+
+    anchor = _namespace["anchor"]
+    ROOT = _namespace["ROOT"]
+    MANIFEST = _namespace["MANIFEST"]
+    ORIENTATION = _namespace["ORIENTATION"]
+    BOOTSTRAP = _bootstrap

odibi_anchor-0.1.0/.assistant/references/INDEX.md

@@ -0,0 +1,73 @@
+# Engineering Reference Library
+
+This is Odibi Anchor's offline technical knowledge layer. Skills prescribe **how to work**;
+references explain **the technology being worked on**. `registry.json` is the closed discovery
+contract. Use `anchor("references")` to list, `anchor("references", "match", "query")` to find, and
+`anchor("references", "load", "id")` to load full content. Task routing returns compact recommendations;
+it never injects full documents or creates a compliance obligation.
+
+For authoritative offline depth, search the licensed snapshots and load only the section needed:
+
+```python
+matches = anchor(
+    "references", "search", "interval selection resolve union",
+    reference_id="visualization.vega-lite", limit=5, output_format="dict",
+)
+section = anchor(
+    "references", "load-section", matches["sections"][0]["section_id"],
+    output_format="dict",
+)
+```
+
+`search` returns bounded previews, source revision, path/lines, and file/section digests without full
+content. `load-section` returns one exact section capped at 32 KiB/200 lines. Search can be global or
+scoped to a registry ID. Both operate entirely from the distributed manifest and snapshots; they do
+not access the network, execute retained source, or write a cache.
+
+## Authored catalog
+
+- **Stack and APIs:** `visualization.stack`, `visualization.api-crosswalk`
+- **Interaction:** `visualization.interactions`, plus the Vega/Vega-Lite/Altair guides
+- **Power BI:** `visualization.deneb-powerbi`, `visualization.powerbi-portability`
+- **Rendering/export:** `visualization.vl-convert`
+- **Typed contracts:** `python.pydantic-v2`, `python.pydantic-contracts`
+
+Load an authored guide first for decisions and boundaries, then search/load source sections for exact
+signatures or schema behavior. A search match is evidence discovery, not proof that the section was
+read. Source inspection does not prove browser, Databricks, or Power BI host behavior.
+
+## Evidence labels
+
+- **[D] Documented:** supported by an authoritative public document.
+- **[S] Source-inspected:** checked against public source or a published schema.
+- **[V] Runtime-verified:** exercised at the stated version in a retained runtime.
+- **[H] Host-verified:** exercised in the named external host.
+- **[U] Unverified:** not exercised in the required runtime or host; treat as a qualification item.
+
+Absence of [V] or [H] is deliberate. Documentation evidence does not prove runtime or host behavior.
+
+## Offline source snapshots
+
+Each registry entry links to licensed, pinned source documentation, schemas, or API files under
+`snapshots/`. `snapshots/manifest.json` records source repository, immutable revision, archive hash,
+file hashes, license, and any source that could not be copied safely. These snapshots are available to
+agents with no network connection; web URLs remain provenance and refresh locations only. Run
+`scripts/update_engineering_reference_snapshots.py` with network access to rebuild the snapshots after
+reviewing a version change. Never update from a floating branch.
+
+## Updating the library
+
+Use the controlled path:
+
+```text
+observation → reference candidate → source/runtime verification → reviewed edit → published reference
+```
+
+Structured learning may identify a candidate but never edits global guidance automatically. Update a
+reference when a registry refresh trigger fires, retain authoritative URLs and exact baselines, label
+unavailable evidence, validate the registry, review the diff, and publish through the ordinary Anchor
+lifecycle. Keep project-specific findings in managed project artifacts rather than this distribution.
+
+Do not copy upstream documentation wholesale. Add original synthesis, decision boundaries, executable
+micro-examples, failure modes, and explicit host limitations. Employer or proprietary material is
+never eligible for this library.

odibi_anchor-0.1.0/.assistant/references/assurance/standards-overlays.md

@@ -0,0 +1,252 @@
+# Assurance standards overlays
+
+- **Reference ID:** `assurance.standards-overlays`
+- **Authored-reference version:** `1.0.0`
+- **Catalog baseline:** `1.0.0`
+- **Canonical overlay catalog SHA-256:** `0277bea95ab65a69255794fe9e4d651a3fcf95548b6b980aa63f93de7ccc473b`
+- **Combined core and overlay catalog SHA-256:** `f58e6ca37d29134e0ace3f94b96315746aa1fd324a1cd120e5f5da3e0cba5d8c`
+- **Captured and last reviewed:** 2026-08-20
+- **Verification label:** `documented`
+
+This Odibi Anchor (Anchor) reference defines eight conditional assurance overlays. It is
+original Anchor guidance informed by the sources in the provenance registry below; it is not an
+imported checklist. An overlay contributes result and evidence expectations only after its
+structured predicate is met. It grants no effect, changes no authority, selects no native skill,
+and establishes no external-framework status.
+
+Repository authority and executable checks outrank this reference. Applicable law, contract,
+explicit owner constraints, effect controls, and safety gates remain authoritative and
+stricter-wins. A reviewed repository exception may strengthen an expectation or explain why a
+control does not concern a change, but it cannot weaken effect safety, evidence truthfulness, or
+an explicit owner requirement.
+
+## Applicability contract
+
+Selection uses normalized structured facts only: work type, execution mode, assurance tier,
+domains, traits, already-authorized declared effects, quality attributes, reviewed changed-path
+classes, and repository capabilities. Free text, filenames alone, agent identity, and skill
+selection are not predicate inputs. A path class must derive from normalized repository-relative
+paths plus inspected file type or content signals; suffix alone cannot activate a high-risk
+overlay. Selection never authorizes an effect.
+
+Within a predicate, `ANY(field,{values})` means that the field's normalized set intersects the
+listed set. `AND` and `OR` have their ordinary Boolean meanings. A selected overlay contributes
+all three controls. Positive reasons use `domain:<value>`, `trait:<value>`, `effect:<value>`,
+`path_class:<value>`, and `tier:<value>`; false predicates emit `predicate_not_met` and no
+controls. Ordering is lexically stable.
+
+### `security.application/1.0.0`
+
+**Predicate:** `ANY(domains,{application,web,api,identity}) OR
+ANY(traits,{handles_secrets,handles_personal_data,authentication,authorization,untrusted_input,network_exposed,cryptography})
+OR ANY(declared_effects,{external_publish,permission_change,secret_change}) OR
+ANY(changed_path_classes,{auth_boundary,security_policy,request_handler,secret_configuration})`
+
+**Sources:** `nist.ssdf`, `owasp.asvs`
+
+- `SEC-INPUT-01/1.0.0` — Untrusted input is constrained at its trust boundary and failure
+  behavior is tested.
+- `SEC-AUTH-01/1.0.0` — Identity and authorization decisions deny unintended access in retained
+  positive and negative tests.
+- `SEC-SECRET-01/1.0.0` — Secrets are neither embedded nor exposed and changed secret handling
+  has a revocation-safe path.
+
+### `data.quality/1.0.0`
+
+**Predicate:** `ANY(domains,{data,analytics}) OR
+ANY(traits,{data_contract,data_transform,data_join,data_write,schema_change,personal_data}) OR
+ANY(declared_effects,{data_mutation,schema_mutation}) OR
+ANY(changed_path_classes,{schema,model,migration,pipeline,quality_rule})`
+
+**Sources:** `iso.25012`, `uk.data-quality`
+
+- `DAT-GRAIN-01/1.0.0` — The output grain and keys are explicit and checked for null or
+  duplicate violations.
+- `DAT-VALID-01/1.0.0` — Critical values are checked against owned validity, completeness,
+  consistency, and timeliness rules.
+- `DAT-TRACE-01/1.0.0` — Material transformations and source-to-output count differences are
+  attributable and unexplained loss is surfaced.
+
+### `ai.agent-risk/1.0.0`
+
+**Predicate:** `ANY(domains,{ai,ml,agent}) OR
+ANY(traits,{model_inference,prompt_or_context_change,tool_calling,retrieval,autonomous_action,human_facing_ai_output,sensitive_model_input})
+OR ANY(changed_path_classes,{prompt,model_config,agent_policy,tool_boundary,retrieval_policy})`
+
+**Sources:** `nist.ai-rmf`, `owasp.genai`
+
+- `AIR-BOUNDARY-01/1.0.0` — Model-controlled content is treated as untrusted and cannot expand
+  tool authority or bypass policy.
+- `AIR-EVAL-01/1.0.0` — Material behavior changes have retained task-representative evaluations,
+  including misuse or failure cases.
+- `AIR-HUMAN-01/1.0.0` — Consequential outputs expose limitations and retain the required human
+  decision or safe-stop boundary.
+
+### `accessibility.user-interface/1.0.0`
+
+**Predicate:** `ANY(domains,{frontend,user_interface,documentation}) AND
+(ANY(traits,{user_facing_interaction,visual_content,audio_video,document_export}) OR
+ANY(changed_path_classes,{ui_component,style,template,user_document}))`
+
+**Sources:** `w3c.wcag`
+
+- `A11Y-OPERATE-01/1.0.0` — Changed user interactions are operable by keyboard with visible
+  focus and no keyboard trap.
+- `A11Y-PERCEIVE-01/1.0.0` — Changed non-text content, structure, status, and contrast have
+  programmatic or textual equivalents appropriate to the medium.
+- `A11Y-ROBUST-01/1.0.0` — Changed interfaces retain meaningful names, roles, values, labels,
+  and error identification under supported assistive technology checks.
+
+### `architecture.change/1.0.0`
+
+**Predicate:** `ANY(traits,{new_component,component_boundary_change,public_contract_change,persistence_change,deployment_topology_change,cross_system_dependency})
+OR ANY(declared_effects,{schema_mutation,infrastructure_mutation}) OR
+ANY(changed_path_classes,{architecture_description,public_api,infrastructure,migration})`
+
+**Sources:** `c4.model`, `arc42.template`, `adr.nygard`
+
+- `ARC-CONTEXT-01/1.0.0` — Affected people, external systems, trust boundaries, and
+  system/component responsibilities are represented at the smallest useful level.
+- `ARC-DECISION-01/1.0.0` — A consequential or hard-to-reverse design choice records context,
+  decision, alternatives, consequences, and status.
+- `ARC-CONSIST-01/1.0.0` — Architecture descriptions, accepted decisions, interfaces, and
+  implementation do not materially contradict one another.
+
+An architecture-description path selects this overlay only when the task changes or repairs an
+authoritative architecture description. Incidental architecture vocabulary is not a positive
+fact.
+
+### `operations.reliability/1.0.0`
+
+**Predicate:** `ANY(domains,{operations,reliability}) OR
+ANY(traits,{production_runtime,scheduled_work,service_level,operational_dependency,on_call_change,capacity_change})
+OR ANY(declared_effects,{deployment,infrastructure_mutation,production_configuration}) OR
+ANY(changed_path_classes,{deployment,observability,runbook,scheduler,service_configuration})`
+
+**Sources:** `google.sre`
+
+- `OPS-SLI-01/1.0.0` — A user-relevant success signal and failure condition exist for the
+  changed production behavior.
+- `OPS-FAIL-01/1.0.0` — Expected failure modes have bounded impact, actionable observation, and
+  a tested recovery or safe degradation path.
+- `OPS-RUN-01/1.0.0` — Operator-facing steps name prerequisites, expected result, escalation,
+  verification, and rollback.
+
+### `supply-chain.software/1.0.0`
+
+**Predicate:** `ANY(traits,{dependency_change,build_change,release_change,artifact_production,third_party_action})
+OR ANY(declared_effects,{external_publish,release}) OR
+ANY(changed_path_classes,{dependency_manifest,lockfile,build_pipeline,release_pipeline,artifact_manifest})`
+
+**Sources:** `nist.ssdf`, `slsa`
+
+- `SUP-PROV-01/1.0.0` — Released artifacts are traceable to a specific source revision and
+  declared build process.
+- `SUP-DEP-01/1.0.0` — Added or changed dependencies have reviewed origin, version, license,
+  integrity, compatibility, and known-vulnerability evidence.
+- `SUP-BUILD-01/1.0.0` — The build does not silently consume undeclared mutable inputs and
+  retained metadata identifies material dependencies.
+
+### `change-safety.high-risk/1.0.0`
+
+**Predicate:** `assurance_tier = T3 AND
+(ANY(traits,{safety_critical,regulated_process,irreversible_change,hazardous_process,large_blast_radius,control_logic_change})
+OR ANY(declared_effects,{production_data_destructive,schema_destructive,permission_change,infrastructure_mutation}))`
+
+**Sources:** `iec.fmea`, `iec.hazop`, `osha.moc`
+
+- `HRC-CHANGE-01/1.0.0` — The proposed change identifies owner, reason, affected baseline,
+  prerequisites, authorization boundary, communication, verification, and restoration path.
+- `HRC-FAILURE-01/1.0.0` — Credible failure modes identify local effect, wider consequence,
+  existing controls, detection, and an owned treatment without converting ordinal judgments
+  into false precision.
+- `HRC-DEVIATION-01/1.0.0` — For hazardous process or control-logic changes, a qualified review
+  considers deviations from intended parameters, causes, consequences, safeguards, and actions
+  before execution.
+
+Selection requires both `tier:T3` and at least one consequence trait or declared consequence
+effect; both reasons remain visible.
+
+## Result evidence
+
+Every applicable control has exactly one assessment state:
+
+| State | Meaning | Shadow projection |
+| --- | --- | --- |
+| `satisfied` | Every required result is present, current, scoped to the change, and passing. | pass |
+| `failed` | A required result ran and shows that the control outcome is not met. | finding |
+| `unavailable` | A required tool, host, permission, fixture, or result is absent, stale, unsupported, or scoped elsewhere. | open uncertainty |
+| `not_assessed` | Applicable evidence has not yet been attempted. | open obligation |
+| `not_applicable` | The predicate is false, or a reviewed structured decision proves that the control does not concern the change. | no obligation |
+| `excepted` | A governed exception identifies control/version, rationale, owner, scope, expiry, and compensating evidence. | visible open risk |
+
+Invocation, elapsed time, an unparsed zero exit status, prose assertions, and loading this
+reference are not result evidence. An evidence record identifies producer, result kind,
+target/scope, capture time, tool or source version, immutable digest, status, and limitations.
+Scope-mismatched or stale evidence is `unavailable`, never `satisfied`. Requirements are
+conjunctive unless a catalog version explicitly names typed, semantically equivalent
+alternatives. Aggregate counts are presentation only; each failure stays visible.
+
+### Evidence required by control family
+
+- `SEC-*`: scoped threat/trust-boundary review and executable negative tests or a relevant
+  scanner result; `SEC-SECRET-01` also requires secret-leak inspection.
+- `DAT-*`: declared contract and grain plus result-backed schema, key, business-rule, freshness,
+  or reconciliation checks as applicable. Profiling cannot establish an owned business rule.
+- `AIR-*`: authority/data-flow review plus retained evaluation cases and observed outcomes.
+  Policy prose and provider assertions are insufficient.
+- `A11Y-*`: an automated accessibility result plus retained manual keyboard and semantic checks
+  for the changed interaction. Automation alone cannot satisfy all three controls.
+- `ARC-*`: inspected architecture or decision artifact plus implementation/interface consistency
+  review. Diagram existence alone is insufficient.
+- `OPS-*`: observed test, simulation, or telemetry result and inspected runbook or rollback
+  evidence as applicable. An unexercised procedure is `not_assessed`.
+- `SUP-*`: dependency/build metadata, integrity/provenance result, and vulnerability/license
+  review scoped to changed inputs. Package-manager success alone is insufficient.
+- `HRC-*`: approved change record, multidisciplinary hazard review where applicable, and retained
+  precondition, verification, and rollback evidence. Tool-generated risk numbers are
+  insufficient.
+
+## Source, license, and refresh provenance
+
+`Baseline` identifies material used to derive original Anchor wording, not a target status. URLs are
+provenance and refresh locations; selection performs no runtime network access. Every source was
+captured for this authored-reference version on 2026-08-20. No paid source snapshot or restricted
+source body is distributed.
+
+| Source ID | Baseline and provenance URL | License and use boundary | Capture date | Refresh trigger |
+| --- | --- | --- | --- | --- |
+| `nist.ssdf` | NIST SP 800-218, SSDF v1.1, February 2022 — https://csrc.nist.gov/pubs/sp/800/218/final | US Government work; cite NIST and use concepts only. | 2026-08-20 | NIST publishes a revision, errata, or successor. |
+| `owasp.asvs` | OWASP ASVS 5.0.0, May 2025 — https://owasp.org/www-project-application-security-verification-standard/ | Upstream CC BY-SA 4.0; preserve attribution while keeping Anchor wording original. | 2026-08-20 | The adopted ASVS major/minor version or license changes. |
+| `iso.25012` | ISO/IEC 25012:2008, Data quality model — https://www.iso.org/standard/35736.html | Copyright ISO; retain bibliographic metadata and public concepts only, with no clauses or tables. | 2026-08-20 | ISO publishes an amendment, revision, or withdrawal. |
+| `uk.data-quality` | UK Government Data Quality Framework, December 2020 — https://www.gov.uk/government/publications/the-government-data-quality-framework | Open Government Licence v3.0; attribute the source. | 2026-08-20 | GOV.UK marks it updated or publishes a successor. |
+| `nist.ai-rmf` | NIST AI 100-1, AI RMF 1.0, January 2023, and NIST AI 600-1 Generative AI Profile, July 2024 — https://www.nist.gov/itl/ai-risk-management-framework | US Government works; cite NIST. | 2026-08-20 | NIST revises the RMF/profile or publishes applicable agent guidance. |
+| `owasp.genai` | OWASP Top 10 for LLM Applications 2025 — https://genai.owasp.org/llm-top-10/ | Upstream CC BY-SA 4.0; original Anchor synthesis with attribution. | 2026-08-20 | OWASP publishes a new adopted list/version or changes scope/license. |
+| `w3c.wcag` | WCAG 2.2, W3C Recommendation, 5 October 2023 — https://www.w3.org/TR/WCAG22/ | W3C Document License; link and paraphrase without republishing normative text. | 2026-08-20 | The Recommendation, errata, techniques, or supported-platform policy changes. |
+| `c4.model` | C4 model website baseline captured 2026-08-20 — https://c4model.com/ | Copyright Simon Brown; site content CC BY 4.0. Attribute and use concepts only. | 2026-08-20 | Site version/license or adopted notation guidance changes. |
+| `arc42.template` | arc42 template v8.2 — https://arc42.org/ | CC BY-SA 4.0; attribute without importing the template wholesale. | 2026-08-20 | The adopted arc42 release or license changes. |
+| `adr.nygard` | Michael Nygard, “Documenting Architecture Decisions,” 2011 — https://cognitect.com/blog/2011/11/15/documenting-architecture-decisions | Copyright source article; use the ADR concept with original Anchor template wording only. | 2026-08-20 | The canonical source changes or the project adopts another ADR contract. |
+| `google.sre` | Google SRE Books online baseline captured 2026-08-20 — https://sre.google/books/ | Book text CC BY-NC-ND 4.0; link and use general concepts without copying or adapting text. | 2026-08-20 | License, canonical guidance, or adopted reliability policy changes. |
+| `slsa` | SLSA specification v1.1 — https://slsa.dev/spec/v1.1/ | Community Specification License 1.0; cite and use provenance concepts without external-status claims. | 2026-08-20 | The adopted SLSA release, track model, or license changes. |
+| `iec.fmea` | IEC 60812:2018, FMEA/FMECA — https://webstore.iec.ch/en/publication/26359 | Copyright IEC; metadata and general method concepts only, with no tables, scales, or clauses. | 2026-08-20 | IEC publishes an amendment, revision, or withdrawal. |
+| `iec.hazop` | IEC 61882:2016, HAZOP studies — https://webstore.iec.ch/en/publication/24321 | Copyright IEC; metadata and general method concepts only, with no guide-word tables or clauses. | 2026-08-20 | IEC publishes an amendment, revision, or withdrawal. |
+| `osha.moc` | OSHA 29 CFR 1910.119(l), Management of change — https://www.osha.gov/laws-regs/regulations/standardnumber/1910/1910.119 | US Government regulation; quote only when legally necessary and separately reviewed. This overlay is not legal advice. | 2026-08-20 | The regulation, an official interpretation, or the governing jurisdiction changes. |
+
+A refresh candidate requires source review, a catalog-version decision, applicable tests, and
+ordinary release review. Learning does not edit this reference or the catalog automatically.
+
+## Limitations and safe use
+
+- The catalog is a bounded assurance aid, not an audit opinion, accreditation, legal
+  determination, or complete framework assessment.
+- Overlay selection does not prove that evidence exists, that a scanner or evaluator is complete,
+  or that a system is safe. Open evidence stays open.
+- The high-risk overlay prompts qualified multidisciplinary review; Anchor does not perform a
+  statutory process-safety study or authorize a hazardous change.
+- The accessibility overlay requires human checks where automation cannot observe the outcome.
+- Security and AI overlays remain catalog/reference concerns; no native skill gains generic
+  security or AI ownership from this reference.
+- Existing skills own procedures. This reference states applicable outcomes and result evidence
+  without replacing data, dependency, documentation, incident, testing, or review workflows.
+- New controls, changed predicates, or changed source baselines require a versioned Spec and
+  review rather than an ad hoc edit.