[chrischeng-c4]

References #75.

Repeated review polling reconstructs deterministic CI prerequisites, while the existing qualifier also mixes those prerequisites with done-label authority. Add one shared read-only readiness result and authenticated receipt interface for later thin consumers, without six controller forks.

Behavior

• Reuse qualification v1 selection and exact latest-attempt snapshots. Extract label-independent evidence without changing valid legacy lifecycle evaluation, failure journals, coding epochs, AGY reclaim or legacy review receipts.
• Add readiness/verify-readiness shared action operations and a separate read-only issue-review verifier entry. Return versioned ready/wait/fail with stable reasons and links. Failure-only does not itself block objective readiness; real qualification blockers still do. Readiness does not grant done, semantic review acceptance or merge authority.
• Bind repo/PR/head/base, complete normalized file inventory, PR description, frozen default-branch policy source/hash, evaluator revision, required checks, workflow/run/attempt/jobs, transitive source pins and nonzero assertion rosters. Authenticate receipts through the configured successful default-branch producer's exact job log, then revalidate current evidence, producer, policy and final PR identity. A hash alone is not authentication.
• One strict JSON object-pairs hook rejects duplicate keys at every depth in metadata, policy/config, wake events, assertion/producer logs and receipt files, including equal values and escape-equivalent names.
• Discover YAML uses from complete PyYAML scanner tokens and BaseLoader representation nodes, including flow mappings. Reject explicit/split/complex/escaped/dynamic/duplicate keys, aliases, tags, directives, multiple documents and resource overflow. No YAML object constructors run. Mutable or undeclared references fail closed; literal block script text is not an action reference.

New parser dependency / execution supply

Readiness only: PyYAML==6.0.3, official release https://pypi.org/project/PyYAML/6.0.3/ . scripts/readiness-requirements.txt locks all 72 official wheel SHA256 values from the release's PyPI JSON metadata; no source builds or transitive packages are accepted. Lock SHA256: eef6794465b73b5dc4ef909500bf5bcc56761b93bb1521a05a97746d1f7f298d.

Only readiness/verify-readiness create a run-local venv below RUNNER_TEMP, install from https://pypi.org/simple using isolated pip, --no-deps --only-binary=:all: --require-hashes, and execute the evaluator in Python isolated mode. The immutable aw reference binds action/code/lockfile. Legacy validate/qualify/handoff skip that bootstrap; main aw package dependencies/uv.lock and permissions are unchanged. No global package or credential change. Missing or different parser versions refuse readiness.

Validation

• Fresh isolated Python 3.13 environment installed PyYAML 6.0.3 through the hash-lock. python -B -m unittest discover -s plugins/faberline/skills/issue-ci/tests -p 'test*.py' -q: 135 passed (89 existing lifecycle/qualification cases plus 46 readiness/distribution/parser cases). Final runtime-scope fixture also passed under -I.
• The flow-adjacent explicit split-key fixture first reproduced the independent review finding, then passed both parser and produce refusal checks. Other negatives cover duplicate JSON, zero/missing assertions, unknown applicability, mutable transitive refs, fake receipts, stale/foreign attempts, producer failure, head/base/policy movement and collection races.
• AST/action-YAML/diff static checks pass. Receipt schema ready/wait/fail fixtures were previously validated and the schema is unchanged by the parser correction.
• Only affected isolated local unit/static fixtures ran. No GitHub workflow, aw cloud check, GKE/e2e/performance execution or manual dispatch was added or run. GitHub runner acceptance of the special YAML forms and runner bootstrap were not exercised; local parser/package-supply checks are not that evidence.

Deliberate first-phase limits

No live consumer/prompt/frequency changes, readiness index publisher, dispatcher/session lease, completion-ledger v2, native webhook/pre-model hook or done activation. Tape #40/#35 still owns the first actual consumer gate; its docs validator is not reimplemented here. Missing index remains an integration blocker, never an empty queue.

Assertion enforcement and actual executed head/base must be established in reviewed real consumer CI; mocked fixtures are not pilot evidence. Pin proof covers declared action/workflow references and enforcement sources, not arbitrary-language imports or a hermetic build. External local-uses checkout context remains unsupported. Verification currently recollects bounded evidence for one named PR; no receipt-only cache or zero-token desktop polling is promised. GitHub has no cross-resource CAS, so consumers must reverify immediately before review/findings.

Draft for independent re-review at 7d5aa3facaa1461c7d22263499a2e2e15b8e9a0e. Do not merge or deploy consumers in this scope.