Fix: Replace Claude model names with OpenCode model IDs in skill frontmatter

[striderZA]

Closes #6

[github-actions]

fatal: could not read Username for 'https://github.com': No such device or address

opencode session  |  github run

[github-actions]

PR Review: fix/model-names

Overall assessment: Good changeset that addresses the immediate issue (#6). The core model name replacements in skills are correct, and the workflow credential fixes resolve the "fatal: could not read Username" error. A few improvements needed.

🔴 Issues

1. Stale documentation in coordination-rules.md (.opencode/docs/coordination-rules.md:21-28)
Still references old Claude model names (model: haiku, model: opus, claude-haiku-4-5-20251001, claude-sonnet-4-6, claude-opus-4-6). Needs updating to match the new OpenCode model IDs.

2. opencode.yml has contents: read but uses persist-credentials: true (.github/workflows/opencode.yml:19)
The review workflow correctly sets contents: write, but opencode.yml keeps contents: read. If opencode ever needs to push changes from comment-triggered workflows, this will fail. Either:

• Add contents: write to opencode.yml for consistency, or
• Revert persist-credentials to false in opencode.yml if it doesn't need to push

⚠️ Warnings

3. Shallow clone may cause issues - Both workflows use default fetch-depth: 1 (shallow clone). If opencode needs full git history (e.g., for diff analysis), add fetch-depth: 0.

4. Missing model: field - Skills like sprint-plan, estimate, retrospective, consistency-check, etc. have no model: field in frontmatter. Not a blocker (they presumably default to something), but worth confirming the default behavior is intentional.

✅ Good

• Correct model mapping: opus → opencode-go/kimi-k2.6, haiku → opencode-go/deepseek-v4-flash
• persist-credentials: true + ${{ secrets.GITHUB_TOKEN }} fixes the auth error in CI
• contents: write in review workflow is appropriate since it needs to post PR comments/reviews
• All 11 skill files were consistently updated

opencode session  |  github run

[github-actions]

PR Review: fix/model-names

Overall: Clean changeset. All 11 skill frontmatter model: fields correctly updated from old Claude model names to OpenCode model IDs. The mapping is consistent: opus → opencode-go/kimi-k2.6, haiku → opencode-go/deepseek-v4-flash. No typos or inconsistencies across files.

✅ Correct mapping verified

Old (Claude)	New (OpenCode)	Skills affected
opus	opencode-go/kimi-k2.6	architecture-review, review-all-gdds, gate-check
haiku	opencode-go/deepseek-v4-flash	changelog, help, onboard, patch-notes, project-stage-detect, scope-check, sprint-status, story-readiness

🔴 Stale reference in coordination-rules.md

.opencode/docs/coordination-rules.md:21-23 still references the old Claude model IDs in the Model Tier Assignment table:

| Haiku  | claude-haiku-4-5-20251001  | ...
| Sonnet | claude-sonnet-4-6         | ...
| Opus   | claude-opus-4-6           | ...

Lines 25-28 also enumerate which skills map to which tier using the old model names. These should be updated to match the new OpenCode model IDs (or remove the table if model IDs are now managed per-skill in frontmatter).

⚠️ opencode.yml already fixed — no issue

The previous review flagged contents: read + persist-credentials: true in opencode.yml, but this is fine. contents: read is sufficient for checkout, and pull-requests: write covers commenting. No change needed.

✅ Other concerns from prior review are non-issues

• Shallow clone: fetch-depth: 1 is standard for CI and fine for opencode review workflows.
• Missing model: on other skills: Out of scope for this PR, which only fixes existing (wrong) model values.

Improvement suggestions

1. Update coordination-rules.md to replace the stale model references with new OpenCode IDs, reflecting what's actually in the skill frontmatter.
2. Cross-check default model behavior: Skills without an explicit model: field (e.g., sprint-plan, estimate) default to something. Confirm the default is the intended model (deepseek-v4-flash presumably) — consider adding a comment or doc note so it's self-documenting.

opencode session  |  github run

[github-actions]

PR Review: fix/model-names

Overall assessment: Clean, consistent changeset. All 11 skill frontmatter model: fields correctly updated. The mapping is uniform: opus → opencode-go/kimi-k2.6 (3 skills), haiku → opencode-go/deepseek-v4-flash (8 skills). No typos or drift.

🔴 Issues

1. Stale coordination rules — docs/coordination-rules.md:21-28 still references claude-haiku-4-5-20251001, claude-sonnet-4-6, claude-opus-4-6, and lists skill assignments using old model names. This will confuse anyone reading the docs. The table and the skill listings on lines 25-28 need updating to match the new model IDs.

✅ What's good

• Consistent mapping applied across all 11 files — no outliers
• Each file touched only the model: line, minimal diff noise
• The Opus-tier skills (architecture-review, review-all-gdds, gate-check) correctly get the stronger model (kimi-k2.6), while Haiku-tier skills get the faster one (deepseek-v4-flash)

Suggestion

Update .opencode/docs/coordination-rules.md to reflect the new model IDs, e.g.:

Tier	Model	When to use
Haiku	opencode-go/deepseek-v4-flash	Read-only status checks, formatting, simple lookups
Sonnet	(default/unset)	Implementation, design authoring, analysis
Opus	opencode-go/kimi-k2.6	Multi-document synthesis, high-stakes gate verdicts

opencode session  |  github run