feat(triage): use validation_loop.schema for output schema resolution

[ggallen]

Summary

• Moves the output schema from env.runner.FULLSEND_OUTPUT_SCHEMA to the new validation_loop.schema field
• Enables schema resolution via fetchBaseFile when the harness is loaded from a URL (config-driven agent registration, ADR-0058)
• Removes the env: block entirely since FULLSEND_OUTPUT_SCHEMA was its only entry

Context

When a harness is loaded from a URL, ${FULLSEND_DIR} doesn't resolve to a meaningful local path, so the schema file is silently skipped. The new validation_loop.schema field (added in fullsend PR #2851) is resolved using the same fetchBaseFile mechanism as validation_loop.script, which correctly fetches and caches the file from the URL base.

Dependencies

Requires fullsend release with PR #2851 (validation_loop.schema support).

Test plan

• Test via ggallen-sandbox config-driven triage run
• Verify schema is fetched from URL and uploaded to sandbox
• Verify validation loop uses the schema correctly

🤖 Generated with Claude Code

[qodo-code-review]

PR Summary by Qodo

Use validation_loop.schema for triage output schema resolution

🐞 Bug fix ⚙️ Configuration changes 🕐 10-20 Minutes

AI Description

• Move triage output schema config to validation_loop.schema for consistent schema resolution.
• Fix schema loading when the harness is fetched from a URL via fetchBaseFile.
• Remove the env.runner FULLSEND_OUTPUT_SCHEMA block since it is no longer needed.

Diagram

graph TD
  A[/"harness/triage.yaml"/] --> B["Harness loader"] --> C["fetchBaseFile"] --> D[("File cache")] --> E["Validation loop"] --> F["Schema validation"]
  G{{"Harness URL"}} --> B
  subgraph Legend
    direction LR
    _cfg[/"Config file"/] ~~~ _svc["Component"] ~~~ _ext{{"External source"}} ~~~ _cache[("Cache")]
  end

Loading

High-Level Assessment

The following are alternative approaches to this PR:

1. Keep FULLSEND_OUTPUT_SCHEMA but resolve it via fetchBaseFile

• ➕ Avoids requiring fullsend PR #2851 support in configs
• ➕ Keeps schema configuration centralized under env.runner
• ➖ Continues relying on env substitution semantics that break in URL-based harness loading contexts
• ➖ Duplicates resolution logic that validation_loop.schema already standardizes

2. Embed schema inline in config (no external file)

• ➕ Eliminates file fetching/caching concerns entirely
• ➕ Works uniformly for local and URL-loaded harnesses
• ➖ Large/verbose YAML; hard to review and maintain schema changes
• ➖ Schema reuse across harnesses becomes difficult

Recommendation: Prefer the PR’s approach: using validation_loop.schema aligns schema resolution with validation_loop.script and leverages the existing fetchBaseFile + caching path for URL-loaded harnesses. The alternatives either reintroduce URL-resolution pitfalls (env-based) or create maintainability issues (inline schema).

Files changed (1) +1 / -4

Other (1) +1 / -4

triage.yamlDefine output schema via validation_loop.schema and remove env.runner override +1/-4

Define output schema via validation_loop.schema and remove env.runner override

• Adds validation_loop.schema pointing at schemas/triage-result.schema.json so schema resolution uses the same base-file fetching behavior as the validation script. Removes the env.runner.FULLSEND_OUTPUT_SCHEMA block since it was the only env entry and does not resolve correctly for URL-loaded harnesses.

harness/triage.yaml

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 7:20 PM UTC · Completed 7:23 PM UTC
Commit: 4d5fec5 · View workflow run →

[fullsend-ai-review]

Review skipped — this PR is already merged.

The /fs-review command only reviews open pull requests.

_{Posted by fullsend pre-review check}

[qodo-code-review]

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 📎 Requirement gaps (0)

Great, no issues found!

Qodo reviewed your code and found no material issues that require review

[fullsend-ai-retro]

🤖 Finished Retro · ✅ Success · Started 7:23 PM UTC · Completed 7:28 PM UTC
Commit: 4d5fec5 · View workflow run →

[fullsend-ai-review]

Review skipped — this PR is already merged.

The /fs-review command only reviews open pull requests.

_{Posted by fullsend post-review check}

[fullsend-ai-retro]

Retro: PR #9 — feat(triage): use validation_loop.schema for output schema resolution

Timeline: Human-authored PR by ggallen, created at 19:17 UTC and self-merged ~76 seconds later at 19:18 UTC. The change was a small, correct config migration (+1/−4 lines in harness/triage.yaml) moving the output schema from env.runner.FULLSEND_OUTPUT_SCHEMA to validation_loop.schema to fix schema resolution when the harness is loaded from a URL.

What happened with agents:

1. Review agent dispatched at 19:17:53 but the PR was already merged by 19:18:55 — the agent detected this at 19:20:43 and posted "Review skipped — this PR is already merged." However, it still bootstrapped a sandbox and ran the agent for ~2 minutes before completing.
2. Three retro runs were triggered (28541913914, 28541976886, 28542000935), likely from multiple webhook events around the merge.
3. Seven issue_comment workflow runs on the source repo were correctly skipped.

Assessment: The workflow was appropriate for a trivial, correct human change. No agent bugs or quality issues were observed. The inefficiencies identified are all well-covered by existing open issues:

• Review on merged PRs / wasted sandbox compute: Covered by #1439, #2388, #2809, and others.
• Multiple retro runs for the same PR: Covered by #2401.
• Retro running on human PRs with no agent involvement: Covered by #2708.

No new proposals are warranted — all improvement opportunities map to existing tracked issues.