docs: document x.klickd supply chain protocol

[Davincc77]

Summary

Consolidates the x.klickd skill-pack build process — previously scattered across RFC-009, the loader protocol, scripts, and benchmarks — into one NON-NORMATIVE, docs-only spec plus an operator quickstart. No code, schema, SDK, package, or CI file is touched; no existing claim is altered.

Added (2):

• docs/rfcs/chimera/SUPPLY_CHAIN.md — supply-chain build-process spec: §0 claim boundary, §1 build request, §2 gated premium pass, §3 the 18-stage pipeline (each stage labelled tool / manual / planned) + §3.1 required context-graph shape, §4 version lineage, §5 logical diff report, §6 rollback / deprecation / approval-revocation / threat-model sizing, §7 determinism + reproducibility, §8 anti-mirage protocol, §9 benchmark intent, §10 audit-trail index + §10.1 security-internal serial fingerprint (private, not described).
• docs/rfcs/chimera/packs/QUICKSTART.md — operator quickstart (human or agent): 5-step build/audit loop, multi-agent role split, shipped verification commands, truth-boundary warning, pre-done checklist.

Edited (3) — pointer links only: docs/rfcs/RFC-009-chimera-v4.1.md §12, docs/rfcs/chimera/README.md §6, docs/rfcs/chimera/packs/README.md §7.

Claim boundary (explicit)

This PR is NON-NORMATIVE, docs-only. It triggers no release (no tag, no latest, no Zenodo DOI, no IANA action). It makes:

• no automation claim beyond the per-stage staged / planned labels — planned stages are specification only and are not presented as working automation;
• no universal-standard and no automatic GDPR / EU AI Act compliance claim — license/boundary checks are engineering checks, not legal attestations;
• no proven-benchmark-superiority claim without external, reproducible evidence;
• no loaded-executable-skill claim without artifact_loaded = true and sha256_matches_manifest = true per the loader protocol.

The 70–80% automation figure is documented as a design target, not a guarantee.

Testing

Check	Command	Result
Pack verifier (artifacts + SHA-256)	python scripts/verify_xklickd_skill_packs.py verify	OK — 42 verified (8 Lite, 34 Pro), all hash-match manifest
Candidate mapping validator	python scripts/validate_v4_1_candidate_mapping.py	OK — 49 rows parsed, 42 artefacts validated, all checks passed
Supply-chain-relevant tests	pytest tests/test_rfc009_scaffold.py tests/test_v4_1_candidate_mapping.py tests/test_starter_pack_validator.py	70 passed, 1 skipped
Full suite (regression baseline)	pytest tests/ -q	6 failed, 131 passed, 1 skipped, 36 errors — identical to pre-change baseline
Codename leak (public surfaces)	grep -i chimera README.md docs/public/*	None (clean)
Banned public claims in new docs	grep	None unbounded — all matches are explicit negations in §0

Baseline note: the 6 failures + 36 errors are pre-existing and environment-only (klickd not installed editable; cryptography / argon2-cffi absent). They fail identically with and without this change. These docs-only changes introduce zero regressions and the baseline is not masked.

Residual risks

• Pipeline maturity is honestly bounded: stages 5/8/9/13 + audit-index/diff/PII-scan are marked planned; only stage 3 and stage 15 (partial) are backed by shipped tooling. The supply chain is not yet end-to-end executable.
• Internal codename containment is partial by design — it remains in docs/rfcs/ paths, benchmark/test/script identifiers, but is absent from true public surfaces (README, docs/public, package metadata). No mass-rename was attempted (high blast radius on CI/test/script paths + published-evidence reproducibility).
• No new automation added — a build runner for the planned stages is net-new code and should be a separate scoped task.

🤖 Generated with Claude Code