fix(#376): standardize agent output filename to agent-result.json

[fullsend-ai-coder]

Summary

Standardize agent output filenames so all agents write agent-result.json instead of per-agent names. The code agent previously wrote code-result.json and the fix agent wrote fix-result.json, diverging from the agent-result.json convention used by all other agents (triage, review, retro, prioritize, scribe).

Changes:

• Remove FULLSEND_OUTPUT_FILE overrides from harness/code.yaml and harness/fix.yaml — the validator now uses the default agent-result.json
• Update agents/code.md and agents/fix.md to write agent-result.json
• Update post-code and post-fix scripts (source and bundled) to look for agent-result.json
• Remove the result.json fallback from validate-output-schema.sh (added as a workaround in fix: accept result.json as fallback in output validation fullsend#1108)
• Update code-implementation and fix-review skills to reference agent-result.json
• Add test cases verifying fallback removal and default filename behavior

Follow-up

After this lands, a follow-up should revert the fallback added in fullsend-ai/fullsend#1108.

Testing

• bash scripts/validate-output-schema-test.sh — all tests pass including new fallback removal tests
• bash scripts/post-code-test.sh — all tests pass
• bash scripts/post-fix-test.sh — all tests pass
• make check-bundle — bundled scripts match source

---

Closes #376

Post-script verification

• Branch is not main/master (agent/376-standardize-output-filename)
• Secret scan passed (gitleaks — 07d44d7302b801b6024604fdfe3fe8402347c94d..HEAD)
• PR body secret scan passed (gitleaks — no-git)
• Pre-commit hooks passed (authoritative run on runner)
• Tests ran inside sandbox

[rh-hemartin]

/fs-review

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 3:10 PM UTC · Completed 3:23 PM UTC
Commit: 7ada4e0 · View workflow run →

[fullsend-ai-review]

Review

Findings

Medium

• [protected-path] All 12 files in this PR are under protected paths (agents/, harness/, scripts/, skills/). The PR links to issue Standardize agent output filename across all agent schemas #376 and the description explains the rationale for standardizing the output filename convention. Human approval is always required for protected-path changes, regardless of context.

The rename from per-agent filenames (code-result.json, fix-result.json) to the standard agent-result.json is mechanically correct and complete — no stale references remain in any file at the PR head. The FULLSEND_OUTPUT_FILE override removal from harness configs is safe because the validator already defaults to agent-result.json. The fallback removal from validate-output-schema.sh is clean, and the new tests adequately verify both the fallback rejection and the default filename behavior. The scope matches issue #376 exactly with no scope creep.

[rh-hemartin]

I reviewed and found no problems

[rh-hemartin]

/fs-fix rebase and resolve the conflicts

[fullsend-ai-coder]

🤖 Finished Fix · ✅ Success · Started 8:20 AM UTC · Completed 8:31 AM UTC
Commit: b2fedcf · View workflow run →

[fullsend-ai-coder]

🔧 Fix agent — iteration 1 (human-triggered)

Rebased onto origin/main and resolved 5 merge conflicts. Updated 2 test files whose helper functions referenced old filenames. All 3 test suites pass (validate-output-schema-test, post-code-test, post-fix-test). Bundle consistency verified with make check-bundle.

Fixed (2):

1. rebase and resolve merge conflicts: Rebased branch onto origin/main (52 commits ahead). Resolved merge conflicts in 5 files: post-code.sh, post-code.src.sh, post-fix.sh, post-fix.src.sh, and validate-output-schema-test.sh. Conflicts arose because main added FULLSEND_VALIDATED_ITERATION_DIR support (using old filenames) while the PR renamed filenames to agent-result.json. Resolution: kept the FULLSEND_VALIDATED_ITERATION_DIR structure from main but applied the PR's filename standardization (agent-result.json), removing the result.json fallback. For the test file, kept both main's in-progress action tests and the PR's fallback removal tests.
2. stale test helpers after rebase: Updated FULLSEND_VALIDATED_ITERATION_DIR test helper functions in post-code-test.sh and post-fix-test.sh to use agent-result.json instead of code-result.json/fix-result.json, matching the filename standardization in the rebased scripts. Removed result.json fallback test cases since the fallback was removed from the scripts.

Tests: passed

_{Updated by fullsend fix agent}

[fullsend-ai-retro]

🤖 Finished Retro · ✅ Success · Started 9:48 AM UTC · Completed 10:11 AM UTC
Commit: 7985da1 · View workflow run →

[fullsend-ai-retro]

Retro: PR #377 — Standardize agent output filename to agent-result.json

Timeline

1. 2026-05-18 — Issue #376 filed by ralphbean about inconsistent output filenames (code-result.json, fix-result.json vs agent-result.json).
2. 2026-07-22 14:49 — Issue transferred to fullsend-ai/agents; triage agent ran immediately, labeled ready-to-code.
3. 2026-07-22 14:55–15:05 — Code agent produced PR fix(#376): standardize agent output filename to agent-result.json #377: 14 files changed (143+, 161−), mechanical rename across agents, harness configs, scripts, skills, and tests.
4. 2026-07-22 15:08–15:23 — Review agent dispatched via /fs-review. Findings: mechanically correct and complete, no stale references, scope matches issue exactly. Only flag: protected-path requires human approval.
5. 2026-07-23 10:31 — rh-hemartin approved: "I reviewed and found no problems."
6. 2026-07-27 08:19–08:31 — rh-hemartin triggered /fs-fix rebase and resolve the conflicts. Fix agent rebased over 52 commits, resolved 5 merge conflicts, updated 2 test files. Force-pushed new head SHA.
7. 2026-07-27 08:31 — pull_request_target/synchronize dispatch fired → "No stage matched — skipping dispatch." No automated re-review.
8. 2026-07-28 09:45 — PR merged by rh-hemartin carrying the stale approval from 5 days earlier.

Workflow quality

Overall: well-executed. The code agent produced correct, well-scoped changes on the first attempt. The review agent's assessment was accurate and aligned perfectly with the human reviewer — zero delta between their evaluations. The fix agent correctly resolved non-trivial merge conflicts. No duplicate dispatches occurred (investigation confirmed two concurrent fix runs were for different PRs: #377 and #429).

One structural gap identified: The fix agent's force-push introduced conflict resolutions and test updates across 7 files, but these changes were never reviewed — neither by the automated review agent nor by a second human pass. The synchronize event from the force-push did not match any dispatch stage, so no review was triggered. The original approval (against a different commit SHA, 5 days stale) was carried forward to merge.

Existing issue coverage

The post-fix review gap is partially tracked by several existing issues:

• fullsend#897 — proposes re-requesting review from human reviewers via the GitHub API after fix pushes (notification-side). This PR provides supporting evidence: the fix agent's force-push on 2026-07-27 resulted in no reviewer notification and no re-review.
• fullsend#5417 — proposes removing stale labels/approvals on force-push (lifecycle management). This PR demonstrates the risk: a 5-day-stale approval on a different SHA was carried to merge.
• fullsend#827 — the review bot did not revoke its stale APPROVED review after the head SHA changed.

However, none of these issues address the dispatch routing gap: fix-agent-originated synchronize events should route to the review stage to trigger an automated re-review. See proposal below.

Autonomy notes

The review agent and human reviewer were fully aligned on this PR. For mechanical rename PRs, the review agent correctly identified scope compliance, completeness, and safety of the changes. The protected-path flag was the only finding, and it was procedurally appropriate. This is a positive data point for review agent competence on mechanical/rename changes across protected paths.

Proposals filed

• Route fix-agent synchronize events to review stage for automated re-review (in fullsend-ai/fullsend)