[wheakerd]

Scope

Refs #218. The routing-context budget measured only the always-loaded gate (5,569 bytes), while the fourteen bundled skill descriptions, which each host lists on every turn as the selection surface, total 7,250 bytes and were not measured anywhere. Records from 0.22.1 now use schema v4, which adds a descriptionSet (each skill's UTF-8 byte and character counts, the totals, and the same labeled token estimate) and a descriptionComparison against the nearest earlier record that carries a description set, under the same 256-byte or 5% review threshold as the gate. The 0.22.1 record migrates to schema v4 as the description baseline with a null comparison; its gate metrics, lifecycle scenarios, and host evidence are unchanged. The validator reproduces the set from the frontmatter the hosts read and rejects drift; earlier records keep their own schemas. The measurement script reports the description set when it measures the gate.

The budget guide and the runtime-changes policy also record each host's hook-output limit as a dated, verified constraint: Codex limits a model-visible hook-output message to roughly 2,500 tokens by default (raised through additionalContextLimit), and Claude Code caps plain SessionStart stdout at 10,000 characters and replaces longer output with a file reference. Both were verified on 2026-10-30 against the official hooks documentation, read one page at a time. A 1.3 multiplier on the byte estimate gives a conservative token reading, under which the Codex default is the binding limit for the gate.

Repository policy revision 116 records the change; the plugin version and runtime contract digest are unchanged.

Affected files and surfaces

• Shared files: None
• Codex-specific files: None
• Claude Code-specific files: None
• Documentation or validation files: axiom_validation/context_budget.py (schema v4 selection, description measurement, shape and comparison checks); scripts/measure-routing-context.py; evals/context-budget/schema-v4.json (new); evals/context-budget/results/v0.22.1.json (migrated to v4 with the description baseline); evals/context-budget/README.md; docs/maintainers/runtime-changes.md; tests/test_context_budget.py; CHANGELOG.md, evidence/repository-policy-revisions-v1.json, evidence/runtime-identity.json, evidence/release-status.json, docs/runtime-identity.md (revision 116)

Routing and authorization impact

• Route-selection impact: None. No skill text changes.
• Action-authorization impact: None.
• New or changed stop conditions: None.

Documentation

evals/context-budget/README.md (Measurement Boundary; Byte Boundary And Record Ownership) and docs/maintainers/runtime-changes.md (Always-Loaded Context) describe the description set and the host limits. Installation and hook commands are unchanged.

Validation

Command	Result
python3 -B scripts/check-publication.py	Passed on this exact tree: 153 required files, 168 Markdown files, 8 routing-context lifecycle scenarios, 30 immutable external action and image pins
python3 -B -m unittest discover -s tests -p 'test_*.py'	OK in a disposable clone of this tree: 312 tests
git diff --check	Passed
Additional targeted checks	python3 -B scripts/measure-routing-context.py --check: passed; python3 -B scripts/measure-routing-context.py: reports the gate plus 14 descriptions totalling 7,250 bytes, equal to the recorded set. ASCII scan of the changed Markdown and JSON files: passed.

Cross-platform parity

• Shared skill changes were reviewed against Codex and Claude Code, or this change does not affect shared skills.
• Manifest, marketplace, and hook differences remain intentional and documented.
• Optional host-native validation results include the host/tool version, or are marked unavailable.

Unrelated work

• I reviewed the final diff and did not reset, hide, stage, or rewrite unrelated work.
• I did not add generated caches, disposable validation copies, local maintenance notes, or tool output.
• I called out every affected file and any routing, authorization, documentation, or parity impact above.