model-release

Releases a model version: checks approvals, tags code, data and parameters, updates the registry and writes release notes.

SKILL.md

GitHub
<!-- Generated from .claude/skills/model-release/SKILL.md by tools/sync_agents.py. Edit the source, then run the script. -->

/model-release — Model Release

Category: Governance · Primary agent: project-planning-specialist

Releases a model version: checks approvals, tags code, data and parameters, updates the registry and writes release notes.

Arguments: whatever the user wrote after the command name.

Purpose

Release a model version for use: verify that validation, the model-risk decision, the model card and a reproducible package are in place; tag code, data vintage and parameters with hashes; write release notes; update the registry; and notify users. For lego models, the manifest must check cleanly with no stale bricks. The output is a traceable release that anyone can reproduce.

Use it when an approved model or model change goes into production use.

Do not use it when approval has not been given (/model-risk-review).

Before you start

  1. Check required inputs. For each input below record FOUND or ABSENT (never 'assumed'):
    • Validation report
    • Model risk decision
    • Model card
    • Replication package If a required input is ABSENT, that part is NOT ASSESSED — NO DATA: do not estimate or invent it; name what is missing and which skill produces it. If all required inputs are absent, stop and say so.
  2. Check confidentiality. Identify the data classification level of the inputs (.claude/docs/confidentiality-and-ai-use.md). Do not proceed with restricted or confidential data in a way the policy does not allow.
  3. Collaboration. Follow .claude/docs/collaboration-protocol.md: ask questions first, present options, get approval before writing files, and update production/session-state/active.md after each approved section.

Agents

  • project-planning-specialist — primary agent; coordinates the work and integrates results
  • mlops-engineer — Engineers MLOps for models
  • reproducibility-engineer — Guarantees reproducibility
  • model-risk-governance-specialist — Defines model risk levels and validation depth, enforces builder-validator independence, governs AI/LLM use and recommends release or block

Each agent is defined in .claude/agents/<name>.md. If your tool runs subagents, run these agents as subagents, in parallel where the steps are independent; otherwise read each agent's file and apply its instructions in turn. Subagents only write to the paths named in your instructions.

Workflow

  1. Verify approvals: validation, model risk decision, model card, replication
  2. Tag code, data vintage and parameter files; record hashes in the registry
  3. Write release notes with changes and known limitations
    • Include changes in outputs relative to the previous version
  4. Update production/model-registry.md; notify users per communication plan
  5. If the model has a lego manifest, require a passing /lego-check, no stale bricks, and record the manifest revision in the release notes

Checkpoints with the user

  • Before tagging: confirm the version number and scope
  • Before notification: confirm release notes

Common pitfalls

  • Releasing without a recorded data vintage
  • Release notes that omit output changes
  • Stale bricks shipped as validated

Output

production/releases/<model>-<version>.md

The release record lists approvals, tags and hashes, changes, output differences from the previous version, limitations, the manifest revision for lego models, and notifications sent.

Gate / verdict

  • No release without a model-risk approval and a reproducible package
  • Output differences from the previous version are documented

Suggested next steps

  • /retrospective