model-migration
Changes a registry model ID across the workspace in one guided pass: edit the SSOT (docs/workspace-schema.json models.*), sweep the sanctioned prose surfaces with sync-model-registry.ts, and verify. Use when: a model is bumped or renamed ("change model", "model bump"), a tier mapping moves to a new model ID, or a model migration needs to reach every sanctioned surface.
- License 许可
- AGPL-3.0
- 最近更新
- 2026-11-14
- Stars
- ★ 25
- Forks
- 9
01 前言 Front Matter
- name
- model-migration
- owner
- lifecycle-manager
- scope
- workspace
- status
- active
- version
- 1.0.0
- metadata
- { "type": "process", "triggers": [ "model-migration", "change model", "model bump", "`모델 변경`" ] }
- description
- Changes a registry model ID across the workspace in one guided pass: edit the SSOT (docs/workspace-schema.json models.*), sweep the sanctioned prose surfaces with sync-model-registry.ts, and verify. Use when: a model is bumped or renamed ("change model", "model bump"), a tier mapping moves to a new model ID, or a model migration needs to reach every sanctioned surface.
- l2_propagate
- false
- last_reviewed
- 1791676800000
- prerequisites
- []
03 SKILL.md 正文
Context
Registry model IDs are registered ONLY in docs/workspace-schema.json (models.<platform>.<tier>) and narrated ONLY in the CLAUDE.md §6 prose and the AGENTS.md §3.6 tier-model-mapping managed block. Literal model IDs are forbidden in skills — skill-lifecycle-audit Check MM fails any skill tree containing a versioned model-ID literal, so skill documents carry registry pointers (docs/workspace-schema.json → models.claude.high) or tier aliases (opus / sonnet / haiku) instead. Historical records (CHANGELOG.md, docs/adr/, docs/designs/, docs/lifecycle/**, registry history columns) deliberately keep the IDs that were current when they were written — the sweep never touches them.
Execution Steps
- Edit
docs/workspace-schema.json— setmodels.<platform>.<tier>to the new registry ID. Caution: the sweep in step 2 refuses to run when--fromis no longer registered anywhere undermodels.*(stale-sweep guard), so if your edit removes the old ID's last registration, run step 2 before saving this one. - Run
bun scripts/sync-model-registry.ts --from <old-id> --to <new-id>. It validates--toagainst the schema, sweeps the curated sanctioned prose surfaces (instruction twins, AGENTS.md rosters, constitution, common and variant docs, and thetests/unit/managed-block-parity.test.tstier-list expectation), prints a per-file replaced-count table, and FAILS if any skills tree still contains a model-ID literal (Check MM pattern set). - Verify:
bun scripts/validate-model-registry.ts(agent frontmatter + tier prose vs registry) andbun test tests/unit/managed-block-parity.test.ts(managed-block tier-list parity). Both must pass before anything is committed. - Record the change per ADR-0074: design doc at
docs/designs/<spec-id>-design.md, thenbun scripts/spec-register.ts --file <design-doc> --source manual --status implemented, plus a CHANGELOG.md entry naming the old and new IDs. - Run
/sync— the L0→L1 propagation publishes the edited twins, rosters, and registry totemplates/common, and the platform mirrors follow.
Related
scripts/sync-model-registry.ts— the sweep tool (curatedSURFACESlist lives there)scripts/validate-model-registry.ts— post-sweep verificationscripts/lib/model-literal-patterns.ts— shared model-ID detection patterns- skill-lifecycle-audit Check MM — the skills-tree literal ban that keeps this migration necessary
docs/workspace-schema.json— the registry SSOT (models.*)
04 同仓库更多 Skills
i18n-layout
Text layout and encoding guidance: character encoding (UTF-8, legacy Korean code pages, BOM hazards), line endings, RTL/bidi handling, and script-specific font selection (Hangul focus). Use when: encoding is corrupted or mojibake appears, an RTL locale must render correctly, fonts must be chosen for a script, or CRLF/BOM issues surface.
i18n-layout
Text layout and encoding guidance: character encoding (UTF-8, legacy Korean code pages, BOM hazards), line endings, RTL/bidi handling, and script-specific font selection (Hangul focus). Use when: encoding is corrupted or mojibake appears, an RTL locale must render correctly, fonts must be chosen for a script, or CRLF/BOM issues surface.
i18n-layout
Text layout and encoding guidance: character encoding (UTF-8, legacy Korean code pages, BOM hazards), line endings, RTL/bidi handling, and script-specific font selection (Hangul focus). Use when: encoding is corrupted or mojibake appears, an RTL locale must render correctly, fonts must be chosen for a script, or CRLF/BOM issues surface.
i18n-layout
Text layout and encoding guidance: character encoding (UTF-8, legacy Korean code pages, BOM hazards), line endings, RTL/bidi handling, and script-specific font selection (Hangul focus). Use when: encoding is corrupted or mojibake appears, an RTL locale must render correctly, fonts must be chosen for a script, or CRLF/BOM issues surface.
i18n-layout
Text layout and encoding guidance: character encoding (UTF-8, legacy Korean code pages, BOM hazards), line endings, RTL/bidi handling, and script-specific font selection (Hangul focus). Use when: encoding is corrupted or mojibake appears, an RTL locale must render correctly, fonts must be chosen for a script, or CRLF/BOM issues surface.
i18n-layout
Text layout and encoding guidance: character encoding (UTF-8, legacy Korean code pages, BOM hazards), line endings, RTL/bidi handling, and script-specific font selection (Hangul focus). Use when: encoding is corrupted or mojibake appears, an RTL locale must render correctly, fonts must be chosen for a script, or CRLF/BOM issues surface.