Practice skill currency: client 2.1.295 to 2.1.300, a source-cited delta of the slots the releases touch - #1171

Scope

• What this PR changes, in one or two sentences: The Claude native-practice skill moves from client 2.1.295 to 2.1.300. A dated record assigns every CHANGELOG entry of 2.1.296, 2.1.299 and 2.1.300 to one category, applies the 13 entries that touch a slot to six rows as evidence and notes, and changes no default. It supersedes Bring the Claude native-practice skill from client 2.1.295 to 2.1.296 with a source-cited delta (CPA-V-09) #1053, which reached 2.1.296 only; the command center ruled on 2026-10-15 that Bring the Claude native-practice skill from client 2.1.295 to 2.1.296 with a source-cited delta (CPA-V-09) #1053 is superseded by one currency change to 2.1.300.
• Base commit: 8053f14e7
• Lane: lane:foundation
• Owned paths touched:
  • docs/decisions/2026-10-15-practice-currency-2300.md (new);
  • tests/test_native_practice_currency.py (new, ported from Bring the Claude native-practice skill from client 2.1.295 to 2.1.296 with a source-cited delta (CPA-V-09) #1053);
  • the practice skill's reference/: slots.json, slots.md, cross-client.md and the 11 layer pages;
  • the slot snapshot manifest (sha256_when_labeled and the rows whose evidence differs from the snapshot);
  • manifests/evidence.json.

The record (anthropics/claude-code@94506ff0 CHANGELOG.md):

There is no 2.1.297 or 2.1.298 section. #1053's 2.1.296 section is byte-identical here, 284 lines lower.

Rows changed (evidence and a note, the same in slots.json and on the page; no default changes):

• context-compaction: L290, L97, L98;
• background-workflows: L291, L245, L65;
• mcp-config-tool-search: L349 (the knob at L1806), L257;
• hooks: L262, L263, L69;
• settings-permissions: L253;
• checkpoints-rewind: L66.

All 13 surfaces are stamped 2.1.300, checked 2026-10-15. Older 2.1.295 mentions that are history stay.

Route additions, accepted (the command center's ruling of 2026-10-15T17:17Z, which checked L65 and L97-L98; applied in 0b1c8be9; no default changes):

1. context-compaction: the route names idle compaction (L97-L98) and keeps idleCompaction and CLAUDE_CODE_IDLE_COMPACT_MIN_TOKENS unset. The ruling's edit adds the exception from CC ruling 2026-10-14: set idleCompaction false only in a session that must keep its exact context across an idle wait over about 54 minutes. Gap ledger: native idle compaction, enableArtifact false as a host choice, PR descriptions read back after publishing #1169 records the same keys as default-on.
2. background-workflows: the route names that a carried workflow resumes through the Workflow tool's approval (L65). The ruling's edit concerns whether a fenced -p run's launch-scoped --allowedTools 'Workflow(<name>)' also covers the carried run. The entry does not say, so the route leaves it out and the notes label it an untested boundary.

Overlaps left to the open PRs that change those rows (listed in the record, not applied). The brief's rule is row-level: do not touch rows that open PRs are changing. Its inventory named pages for #1142 and #1146 (workers) and for #1147 (native-clients), so the rows were checked one by one. Each open PR's slots.json was compared row by row with its merge base on 2026-10-15T17:31Z:

• Practice skill: spawn one-shot workers unnamed; shut teammates down when their result is sealed #1142: agent-teams, subagent-definitions;
• Practice rows: --max-findings, marketplace autoUpdate and trust, the workflow model override, the effortLevel origin #1147: model-effort-routing, plugin-marketplaces, pr-review-pre-cue;
• Record the own-clone rule for sandboxed runs and the vendor sandbox's side effects #1129: client-core, sandbox;
• Gap ledger isolation: background sessions isolate before editing, lanes park work as WIP commits, the strict sandbox names its allowlist #1160: sandbox, worktrees;
• Gap ledger: observation-inference (entrypoint label, plan limits) and scheduling-supervision (restored cron tasks) #1158: scheduled-tasks, usage-cost-monitoring;
• Gap ledger agent-sdks: background-command limit, bounded retry watchdog, message accept scope, Agent SDK 0.2.167 #1157: headless-sdk;
• Keep duplicate /skill-doctor rows, refuse zero-row tables, report unused plugins; record the 2026-10-15 plugin dispositions #1153: skills-library;
• Memory maintenance: a read-only index guard reading the client's own limits; auto-dream detection; auto-memory row #1149: auto-memory;
• Currency notice: count a practice skill behind the client and a screened practice reference that changed #1136: practice-references, release-tracking;
• Raise the Claude Code floor to 2.1.300 and record its source commit as the run pin #1128: client-core.

None of the six rows edited here is among them.

• subagent-definitions (Practice skill: spawn one-shot workers unnamed; shut teammates down when their result is sealed #1142): L290, L343, L254;
• model-effort-routing (Practice rows: --max-findings, marketplace autoUpdate and trust, the workflow model override, the effortLevel origin #1147): L291;
• usage-cost-monitoring (Gap ledger: observation-inference (entrypoint label, plan limits) and scheduling-supervision (restored cron tasks) #1158): L6;
• sandbox (Record the own-clone rule for sandboxed runs and the vendor sandbox's side effects #1129, Gap ledger isolation: background sessions isolate before editing, lanes park work as WIP commits, the strict sandbox names its allowlist #1160): L129, L134;
• headless-sdk (Gap ledger agent-sdks: background-command limit, bounded retry watchdog, message accept scope, Agent SDK 0.2.167 #1157): L171.

SOTA sources

• anthropics/claude-code@94506ff CHANGELOG.md, sha256 8c6fdbbd692bd9b2e108e9946948aa77200b6c8ffec4f807ea9c8cee92545139: L3-L92 (2.1.300), L94-L285 (2.1.299) and L287-L367 (2.1.296), plus L1806 and L371. https://github.com/anthropics/claude-code/blob/94506ff0d3bd9185f1752da36e968775021fb48f/CHANGELOG.md
• anthropics/claude-code@3deb873 CHANGELOG.md, sha256 0383b383e05c30b7592c512ea3356104cbc3a799d028d36204033175ff256a51: the version Bring the Claude native-practice skill from client 2.1.295 to 2.1.296 with a source-cited delta (CPA-V-09) #1053 read.
• Bring the Claude native-practice skill from client 2.1.295 to 2.1.296 with a source-cited delta (CPA-V-09) #1053 at 213a9b4d8: docs/decisions/2026-10-12-practice-currency-2296.md and tests/test_native_practice_currency.py, superseded here.
• The skill's own rule, .claude/skills/claude-native-practice/SKILL.md step 2: read the CHANGELOG between the page's version and the installed client.

Evidence-class table

Local commands run

$ python3 -m unittest tests.test_native_practice_currency   # git archive export of e1b528805 with the new test and record absent from the base
Ran 10; FAILED (failures=21, counting 13 page subtests), e.g. '2.1.295' != '2.1.300'; '0 != 13' applied entries
$ python3 -m unittest tests.test_native_practice_* tests.test_evidence_manifest tests.test_validate tests.test_host_receipts tests.test_published_digest_privacy tests.test_blind_checkout tests.test_local_pages_overview tests.test_paired_runs_results tests.test_primary_source_reading_counts tests.test_resolution_protocol_docs tests.test_skills_manifest tests.test_skills_status   # f8a49399, rebased on 8053f14e7
Ran 678 tests; OK (skipped=6)
$ python3 -m unittest tests.test_native_practice_currency.DeltaTests.test_the_accepted_route_additions_are_on_the_rows_with_the_rulings_edits   # git archive export of f8a49399a with the new test
FAILED: 'accepted both' not found in the record
$ python3 -m unittest tests.test_native_practice_* tests.test_evidence_manifest tests.test_validate tests.test_host_receipts tests.test_published_digest_privacy tests.test_skills_manifest tests.test_skills_status tests.test_local_pages_overview   # 0b1c8be9
Ran 580 tests; OK (skipped=4)
$ python3 -m unittest tests.test_native_practice_currency.DeltaTests.test_every_page_evidence_line_is_the_join_of_its_slots_json_list   # git archive export of 0b1c8be98 with the new test
FAILED for background-workflows only (the CC's MICRO read, P2: the page line had split the paired-run item)
$ (the same 581-test reader set)   # 2d6f832d
Ran 581 tests; OK (skipped=4)
$ python3 scripts/validate.py
rc 0; {"components": 71, "hashed_files": 12141, "profiles": 4, "receipts": 240, "status": "passed"}

Before the rebase, on e1b528805, the builder also ran the 13 modules that walk every tracked file: 30 modules, 1,502 tests, 0 failures. #1161's owner-quote detector finds 0 hits in the changed files.

Decision record

docs/decisions/2026-10-15-practice-currency-2300.md. Its evidence is the pinned CHANGELOG. It considered keeping #1053's 2.1.296-only record, which is superseded. The two route additions follow the command center's ruling of 2026-10-15T17:17Z. A row's own overturn_when names the comparison that would change it.

Follow-ups

• The hooks note says the pinned floor is 2.1.284, while adoption/pins-linux-x86_64.json says 2.1.295. That sentence is outside this delta.
• workers.md's background-workflows evidence line already carries two 12-hex prefixes on main, bound in the committed evidence/artifacts/native-practice-paired-runs-20261010/paired-runs.json. Appending this PR's evidence re-shows that line in the diff, and this PR adds no digest of its own. In slots.json the new items go before that line, so it is unchanged there.
• When Re-baseline the upstream surface watch at Claude Code 2.1.300; hold cached-artifact checks to the baseline #1155 lands, the version test becomes plain equality with the baseline, with no edit needed.
• The pure-bug-fix and out-of-scope assignments rest on the entry text. The hooks fail-closed cases were not re-run on 2.1.300.

Checklist

• No GitHub Actions changed.
• No workflows changed.
• No secrets are printed, logged or committed.
• No new paid hosting, subscription or billing surface.
• Peer-owned untracked files and worktrees were preserved.

🤖 Generated with Claude Code