feat(scaffold): add role and slug to harness templates (ADR-0045 PR 6)

[ggallen]

Summary

• Adds role and slug fields to all 6 scaffold harness templates in internal/scaffold/fullsend-repo/harness/
• Part of ADR-0045 Phase 1 (implementation plan, PR 6)
• Depends on PR 2 (feat(harness): add optional role and slug fields (ADR-0045 PR 2) #2128, merged) which added optional role/slug fields to the Harness struct

Harness → role/slug mapping

Harness	Role	Slug
triage.yaml	triage	fullsend-ai-triage
code.yaml	coder	fullsend-ai-coder
review.yaml	review	fullsend-ai-review
fix.yaml	coder	fullsend-ai-coder
retro.yaml	retro	fullsend-ai-retro
prioritize.yaml	prioritize	fullsend-ai-prioritize

Note: fix.yaml reuses the coder app (same role/slug as code.yaml).

Test plan

• make go-test — all tests pass, including TestHarnessesLoadAndValidate
• make lint — passes
• Backward compatible — fields are omitempty, existing harnesses without role/slug continue to work

🤖 Generated with Claude Code

[github-actions]

Site preview

Preview: https://c6fbacd3-site.fullsend-ai.workers.dev

Commit: 7ac6d8ec3f53cb6b47c0cb430deb231ba2685c35

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 8:57 PM UTC · Completed 9:04 PM UTC
Commit: b02a3a1 · View workflow run →

[fullsend-ai-review]

Review

Findings

Info

• [authorization-implicit] internal/scaffold/fullsend-repo/harness/code.yaml — No linked issue exists. Authorization is inferred from: (1) ADR-0045 has Accepted status, (2) the implementation plan explicitly specifies this as PR 6, (3) the change is mechanical (18 lines adding pre-defined values across 6 templates), and (4) the prerequisite PR feat(harness): add optional role and slug fields (ADR-0045 PR 2) #2128 (which added the schema fields) is merged. All 6 role/slug values match the implementation plan specification exactly.

Previous run

Review

Findings

Low

• [field-ordering-inconsistency] internal/scaffold/fullsend-repo/harness/code.yaml — All 6 scaffold harness templates place role and slug immediately after agent and before doc, while the ADR-0045 "Full example" (line 283-290) places them after agent, model, image, and policy with a blank-line separator. However, the Go struct (harness.go:196-200) orders Role/Slug right after Agent/Doc/Description, which is closer to the PR's ordering. The two authoritative sources disagree on canonical ordering, and YAML key order has no functional impact. This is a minor style inconsistency across all 6 files: code.yaml, fix.yaml, prioritize.yaml, retro.yaml, review.yaml, triage.yaml.
  Remediation: Pick one ordering convention (ADR example or Go struct) and apply it consistently. The ADR example ordering (role/slug after policy, separated by blank line) is slightly more readable as it groups identity fields visually.

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 9:08 PM UTC · Completed 9:14 PM UTC
Commit: 7ac6d8e · View workflow run →

[fullsend-ai-retro]

🤖 Finished Retro · ✅ Success · Started 11:04 AM UTC · Completed 11:08 AM UTC
Commit: 7ac6d8e · View workflow run →

[fullsend-ai-retro]

Retro: PR #2140 — feat(scaffold): add role and slug to harness templates

This was a clean, well-executed workflow with no meaningful improvement opportunities identified.

Timeline: A mechanical 18-line PR (6 files) was opened by a human author. The review bot completed its first pass in ~8 minutes, finding a single low-severity field-ordering inconsistency between the ADR-0045 example and the PR's YAML key ordering. The bot correctly approved despite the low finding. The author fixed the ordering within 1 minute and force-pushed. The second review pass approved cleanly with no new findings. A human reviewer approved ~9 hours later, and the PR merged shortly after.

Assessment: The review agent performed well — it caught a legitimate style inconsistency, correctly classified it as low severity, provided a clear suggested fix, and approved rather than blocking. The author's fast turnaround minimized rework cost. The sticky comment history tracking worked as designed, collapsing the resolved finding into a "Previous run" block. Total token cost was reasonable (2 review runs for a small PR). No redundant work or dead-end exploration observed.

Existing coverage: Several open issues already address the closest potential improvements:

• Review agent: use COMMENT verdict for re-reviews with only low-severity findings #2029 / Review agent: use COMMENT verdict for human-authored PRs with only medium/low findings #2115 — verdict calibration for low-severity findings
• Review agent surfaces new low-severity findings across multiple approved passes #1367 — low-severity findings across multiple passes
• Extend retro skip guard to human-authored PRs with no-finding reviews #1675 — retro skip guard for clean reviews

No new proposals warranted.