Move the claude-native-practice skill row to the commit that holds #1001, #1005 and #986 - #1018

Scope

• What this PR changes, in one or two sentences: moves the claude-native-practice row of adoption/skills/manifest.json and this repository's source pin in catalogs/landscape/skills-lifecycle.json from 7a6bfbcf0 (Ground the Claude Code native practice in primary sources and one cross-client mapping #958) to c59655c7c85a014b86699240754c382c75a475a6, the commit of main that holds the final state of the skill after Retire the skill's Claude-in-GitHub rows and point them to local review #1001, Record the paired-run results (planning, orchestration) and pin the RTK receipt's locators #1005 and Follow-up to #958: the advisor-spend record's four P3s and the cross-client page's Claude sources #986 landed (the ONE re-pin the command center ruled for the three pull requests), and appends a dated amendment to the row's decision record. Only the measured pin fields change, from the commit's own bytes: url, ref, tree_sha, skill_md_sha256, skill_md_bytes, and the manifest's checked_at (2026-10-10 to 2026-10-11). The description stays 349 characters, so the listing sums and the 11,000 ceiling are unchanged, and every other field of the row (first-party flag, listing, Codex enablement, audit text, gap text) equals the measured one. The skill is not installed by this change.
• Base commit: c59655c7c85a014b86699240754c382c75a475a6 (main at the landing of Follow-up to #958: the advisor-spend record's four P3s and the cross-client page's Claude sources #986, 2026-10-11T20:30:45Z).
• Lane: lane:foundation
• Owned paths touched: adoption/skills/manifest.json, catalogs/landscape/skills-lifecycle.json, docs/decisions/2026-10-10-skills-claude-native-practice-row.md, and the three registrations in manifests/evidence.json.

The row sets of the three pull requests inside the new pin (from git diff of each landing commit against its parent, limited to .claude/skills/claude-native-practice; 9 skill files and 11 slot rows net, usage-cost-monitoring changed in two of them):

SOTA sources

• This repository's own skill at the pinned commit: .claude/skills/claude-native-practice at c59655c7c85a014b86699240754c382c75a475a6 (a commit of main; git merge-base --is-ancestor exit 0, and the skill directory is identical at main's head). Tree f08d0d241b774112bbc1e1f43558e7604e4916b6 (was 3a485d2a4bbe63dc1ae54e23101c472d21450b8c), SKILL.md sha256 005288455c9d6a0bd70772847e2bb2411554d90eeabc0aa135af5d12fe99b353 (2,290 bytes, was 2,224 bytes, a52a24bc7691fa18b3e4c532c6cbd15b79968d82a54acb4202c9fcced2ad8bcb). The values were read from the git objects of a blobless clone of the public remote, which is the verification adoption/skills/lifecycle.md prescribes (git clone --filter=blob:none --no-checkout, then git rev-parse <ref>:<path> and git show <ref>:<path>/SKILL.md).
• The pinned installer is unchanged: the Vercel Skills CLI skills 1.7.0 (vercel-labs/skills v1.7.0, the manifest's cli block) installs a skill from an immutable GitHub tree URL (adoption/skills/lifecycle.md, "Update and recover": review the exact changed skill directory, update its immutable URL, ref, tree and file hashes, then rerun the same installer). The route and its first-party form are recorded in docs/decisions/2026-10-10-skills-claude-native-practice-row.md ("Source and row"), landed by Pin the claude-native-practice skill as a first-party manifest row and raise the listing ceiling to 11,000 #972 (the pull request that added the record and the manifest row; Ground the Claude Code native practice in primary sources and one cross-client mapping #958 landed the skill content the row pins).
• The listing method is unchanged: the manifest's description_chars_method and the repository counter scripts/skills_status.py; the description is the same 349 characters, so no new vendor claim about the listing budget is made here.

Evidence-class table

Local commands run

$ git clone --filter=blob:none --no-checkout https://github.com/seathatflowsinourveins/native-agent-stack.git repo   # fresh directory
$ git rev-parse c59655c7c85a014b86699240754c382c75a475a6:.claude/skills/claude-native-practice
f08d0d241b774112bbc1e1f43558e7604e4916b6
$ git show c59655c7c85a014b86699240754c382c75a475a6:.claude/skills/claude-native-practice/SKILL.md | sha256sum
005288455c9d6a0bd70772847e2bb2411554d90eeabc0aa135af5d12fe99b353  -
$ git show c59655c7c85a014b86699240754c382c75a475a6:.claude/skills/claude-native-practice/SKILL.md | wc -c
2290
$ python3 scripts/validate.py
{"components": 70, "hashed_files": 11670, "profiles": 4, "receipts": 239, "status": "passed"}
$ python3 scripts/evidence_manifest.py --check
{"files": 11670, "status": "passed"}
$ python3 -m unittest <the 26 modules that name or consume the changed files, listed below>
Ran 1416 tests in 668.760s
OK (skipped=14)
(exit 0; run at a host load average of 60 to 100)

The 26 modules: test_skills_manifest, test_skills_status, test_install_skills, test_runtime_worker_skills, test_landscape_sweep_skills, test_new_wsl_client_config, test_evidence_manifest, test_skill_usage, test_native_practice_cross_client_sources, test_native_practice_doc_corrections, test_native_practice_meter_windows, test_native_practice_retired_routes, test_native_practice_snapshot_provenance, test_paired_runs_results, test_primary_source_reading_counts, test_architecture_pin_source, test_bootstrap_full_profile, test_currency_due, test_runtime_worker_openhands, test_runtime_worker_openhands_resolver, test_starred_convergence, and five of test_local_pages_architecture_* (cache, inventory, landscape_exclusions, manifest, read_boundary). The edit was made with a local script that is not part of the repository; the commands above reproduce its result from the git objects.

Decision record

docs/decisions/2026-10-10-skills-claude-native-practice-row.md, new section "Re-pin (2026-10-11)". Alternatives considered: leave the row at 7a6bfbcf0 (then the installed skill would keep the retired Claude-in-GitHub route text of #958 and lack the paired-run rows and the cross-client sources until some later change), or re-pin once per pull request (three moves of the same row; the command center ruled one). What would overturn this: the reversible inverse is to revert this commit, which restores the previous pin and the registry rows; an installed copy follows only through the lifecycle's install or retirement procedure.

Host evidence

Not applicable: this PR adds or changes no file under evidence/hosts/.

• python3 scripts/host_receipts.py validate passes for every new/changed receipt. (not applicable)
• Independent review is recorded (scripts/host_receipts.py review) or explicitly requested in this PR. (not applicable)
• No platform_status change is made from a host receipt alone; a platform status flip still needs the matrix rule in docs/component-evidence-matrix.md. (not applicable)

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 designated host shell installs the skill from the new pin by adoption/skills/lifecycle.md after this lands; an installed copy at the old pin is stale until then (a host action, not part of this change).
• The WS1 and WS3 changes to the slot rows (the dispositions of the claude-practice stars and the research-practice record pointer on the matching worker, SDK and instruction-file rows) edit reference/slots.json, so they move the skill tree again and are followed by their own re-pin; they wait for Finalize 373 starred repositories with guarded retention and source coverage #1016 and docs: consolidate research convergence and close practice workstreams #1011.
• Astra's read, P3-1 (the first read of 83007bfb0 is an ACK with no blocker; its three P3s sit outside the changed lines): the unchanged installer mechanism has no mirror at its pin. The Vercel Skills CLI skills 1.7.0 at 7407f3893ad4dceab546ac002c3ef806e4000c73 (cited by adoption/skills/manifest.json:8, the decision record and adoption/skills/lifecycle.md:284) is not in ~/code/upstream/MIRRORS.json, so the read could not check the immutable-tree-URL and update implementation, the src/cli.ts:398 locator or the package digests against vendor bytes; the installer tests use a simulated CLI. This PR changes no mechanism and no false claim was established. Follow-up: register that pin in the upstream mirror list, verify the locator and the package hashes at it and keep the result.
• Astra's read, P3-2: manifests/evidence.json holds 20 pre-existing occurrences of the host's machine name on 18 lines, none changed by this PR (counted, not printed). Normalizing them is an optional separate change for the evidence owner under the hostname ruling and never blocks this PR.
• Astra's read, P3-3: some tests could not run in the read's sandbox (fail-closed checks reject the packet's enclosing Git repository; others need Git metadata, sockets, native dependencies or a matching installed host artifact); its base and head controls separate these from merge regressions. This body's 26-module run was made from a genuine worktree with scratch outside every repository (1,416 tests OK) and CI at the head passes. Follow-up: the affected commands from genuine exact-revision worktrees, with the results attached; no refusal check is weakened.

🤖 Generated with Claude Code