Bring the Claude native-practice skill from client 2.1.295 to 2.1.296 with a source-cited delta (CPA-V-09) - #1053

Scope

• What this PR changes, in one or two sentences: CPA-V-09 of the converged Claude practice audit. The Claude native-practice skill (its layer pages, the index and slots.json) moves from client 2.1.295 to 2.1.296, checked 2026-10-12, with a source-cited delta for the five rows the release's entries touch and a decision record that assigns all 79 entries of the 2.1.296 section once; no default, route or overturn condition changes.
• Base commit: 703d588c5 (main after Add isolated Pi and OpenCode trial profiles and native invocation receipts #1033).
• Lane: lane:foundation
• Owned paths touched: .claude/skills/claude-native-practice/reference/ (13 pages and slots.json), docs/decisions/ (one new record), evidence/artifacts/claude-native-practice-20261009/claude-native-slot-manifest.json (its label), tests/test_native_practice_currency.py (new), manifests/evidence.json.
• Commits: one.

Why

claude --version prints 2.1.296 (Claude Code) and catalogs/foundation/upstream-surface-baseline.json (versions.claude_code) says 2.1.296, while the skill's pages and slots.json said 2.1.295, checked 2026-10-09. The skill's own rule (SKILL.md step 2) is to read the CHANGELOG entries between the two versions and to treat a row as current only if none changes the behaviour its default depends on; a routing or compaction default changes only after its own paired run. The record is that read.

What it adds

SOTA sources

• anthropics/claude-code@3deb8736 CHANGELOG.md L5-L83 (the 2.1.296 section; L6, L7, L59 and L65 are the entries that touch slots), L1522 (2.1.280: CLAUDE_CODE_MAX_MCP_DESCRIPTION_LENGTH) and L87 (2.1.295: onFailure: "block", the one 2.1.295 fact the pages keep as history).
• This repository: .claude/skills/claude-native-practice/SKILL.md (the currency rule), catalogs/foundation/upstream-surface-baseline.json (versions.claude_code 2.1.296, resolved by 8b1ef0c37), docs/decisions/2026-10-09-claude-code-native-practice.md (the skill's record).
• The converged practice audit, group CPA-V-09 (the command center's ruling of 2026-10-12T13:38:12Z accepted the converged set as the work queue; a coordination record, cited by name and time).

Evidence-class table

Local commands run

$ python3 -m unittest tests.test_native_practice_currency tests.test_native_practice_snapshot_provenance  ->  Ran 16 tests, OK
$ python3 -m unittest <the 58 modules that name a changed file, one process each>  ->  3,676 tests; 57 modules OK; the one failure (snapshot provenance: the label had to follow the new slots.json) is fixed in this commit and passes
$ python3 scripts/validate.py  ->  exit 0;  python3 scripts/evidence_manifest.py --check  ->  exit 0

Decision record

docs/decisions/2026-10-12-practice-currency-2296.md.

Host evidence

Only relevant when this PR adds or changes files under evidence/hosts/. This PR changes none.

Checklist

• New/changed GitHub Actions are pinned to a full commit SHA with a version comment (no floating tags). None changed.
• New/changed workflows declare top-level permissions: {} and grant each job only what it needs. None changed.
• No secrets are printed, logged or committed; no new required secret was added without a documented owner.
• No new paid hosting, subscription or billing surface was introduced.
• Peer-owned untracked files and worktrees were preserved (not deleted, moved or overwritten).

Follow-ups

• The practice skill is installed on the host from this directory; re-installing it (install_skills.py --only claude-native-practice) is the command center's act after landing.
• The Codex column of the cross-client table (0.161.0 there, rust-v0.162.1 in the baseline) is not part of this change.
• CPA-V-10 (a check that detects drift between the installed user agents and the catalog) is a separate change; this one only re-measures the note.

🤖 Generated with Claude Code