fix(#483): defer target_branch until after default branch discovery by fullsend-ai-coder[bot] · Pull Request #488 · fullsend-ai/agents

fullsend-ai-coder · 2026-07-28T09:16:48Z

Summary

Fixes branch validation failures on repos with non-main default branches by:

skills/code-implementation/SKILL.md (step 3): Adding fallback branch discovery commands (git symbolic-ref, gh repo view) so the agent tries multiple methods before writing target_branch. Documents that the post-script auto-corrects when no explicit allowed list is configured.
scripts/post-code.src.sh: When CODE_ALLOWED_TARGET_BRANCHES is unset and the agent's target doesn't match the API-discovered default, auto-correcting to the default branch with a warning instead of hard-failing. Hard failure is preserved when an explicit allowed list is set.
scripts/post-code.sh: Regenerated via make script-build.
scripts/post-code-test.sh: 8 new test cases covering the branch validation decision matrix (auto-correct, explicit list reject, match, wildcard, no-target fallback, etc.).

Testing

All 76 post-code tests pass (68 existing + 8 new)
make check-bundle confirms bundled script is in sync
Pre-existing failure in post-triage-test.sh (prerequisites-creates-allowed-issue) is unrelated to these changes

Closes #483

Post-script verification

Branch is not main/master (agent/483-defer-target-branch)
Secret scan passed (gitleaks — 3760b3bb70de32596c1922332fb96e76cd9ebcd8..HEAD)
PR body secret scan passed (gitleaks — no-git)
Pre-commit hooks passed (authoritative run on runner)
Tests ran inside sandbox

rh-hemartin · 2026-07-28T09:33:14Z

/fs-fix make the command gh repo view --json defaultBranchRef --jq '.defaultBranchRef.name' the first one to be tested and then the other ones.

fullsend-ai-coder · 2026-07-28T09:34:50Z

🤖 Finished Fix · ✅ Success · Started 9:34 AM UTC · Completed 9:39 AM UTC
Commit: de7ace2 · View workflow run →

Reorder branch discovery commands in SKILL.md so `gh repo view` is tried first, followed by `git rev-parse` and `git symbolic-ref` as fallbacks. Addresses review feedback on #488

fullsend-ai-coder · 2026-07-28T09:39:00Z

🔧 Fix agent — iteration 1 (human-triggered)

Reordered branch discovery commands in SKILL.md per human instruction: gh repo view is now the first method tried, with git rev-parse and git symbolic-ref as fallbacks.

Fixed (1):

reorder branch discovery to prioritize gh repo view (skills/code-implementation/SKILL.md): Reordered the branch discovery commands so gh repo view --json defaultBranchRef --jq '.defaultBranchRef.name' is tried first, followed by git rev-parse --abbrev-ref origin/HEAD and git symbolic-ref refs/remotes/origin/HEAD as fallbacks.

Tests: passed

_{Updated by fullsend fix agent}

rh-hemartin · 2026-07-28T09:39:50Z

/fs-review

fullsend-ai-review · 2026-07-28T09:41:18Z

🤖 Finished Review · ✅ Success · Started 9:41 AM UTC · Completed 9:56 AM UTC
Commit: c321286 · View workflow run →

fullsend-ai-review · 2026-07-28T09:56:32Z

Review

Findings

Medium

[protected-path] scripts/post-code.src.sh, scripts/post-code.sh, scripts/post-code-test.sh, skills/code-implementation/SKILL.md — All four changed files are under protected paths (scripts/, skills/). The PR links to issue fix(code): defer target_branch until after default branch discovery #483 and provides clear rationale for the changes. Human approval is always required for protected-path changes, regardless of context.
[behavior-description-incomplete] docs/code.md:44 — The Default column for CODE_ALLOWED_TARGET_BRANCHES describes the default value as "Repo default branch (auto-detected via GitHub API; falls back to main)" but does not explain the new auto-correct behavior. After this PR, when CODE_ALLOWED_TARGET_BRANCHES is unset, the post-script auto-corrects the agent's chosen branch to the default branch rather than validating and rejecting a mismatch. The current description conflates the default-branch discovery mechanism with the variable's behavioral default.
Remediation: Update the Default column to clarify: "When unset: auto-corrects agent's choice to repo default branch. When set: validates agent's choice against the specified allowlist."

Low

[test-integrity] scripts/post-code-test.sh:1107 — Tests reimplement the branch validation logic in a standalone function (validate_target_branch) rather than exercising the production code. The test helper faithfully mirrors the production logic at PR head, but logic drift between the two would not be caught. This matches existing test patterns in the file, so severity remains low.

Previous run

Review

Findings

Medium

[protected-path] scripts/post-code.src.sh, scripts/post-code.sh, scripts/post-code-test.sh, skills/code-implementation/SKILL.md — All four changed files are under protected paths (scripts/, skills/). The PR links to issue fix(code): defer target_branch until after default branch discovery #483 and provides clear rationale for the changes. Human approval is always required for protected-path changes, regardless of context.

Low

[authorization-change] scripts/post-code.src.sh:135 — When CODE_ALLOWED_TARGET_BRANCHES is unset, the behavior changed from fail-closed (hard-fail on mismatch) to auto-correct (silently use the API-discovered default branch). This is the intended fix per issue fix(code): defer target_branch until after default branch discovery #483 and the auto-corrected value is always the safe default. Operators who relied on the old implicit "reject mismatches" behavior should be aware of the change.
[documentation-coherence] skills/code-implementation/SKILL.md:204 — The bolded text says "Do not hardcode main as a fallback" but the jq template two lines later uses ${DEFAULT_BRANCH:-main}, which falls back to main. Consider rewording to "Do not skip discovery and assume main" to avoid confusing the agent.
[edge-case] scripts/post-code-test.sh:1139 — Consider adding a test for substring matching (e.g., agent_target='release', allowed_list='release-1,release-2'). The comma-wrapping technique correctly rejects partial matches; codifying this in a test guards against regressions.
[test-integrity] scripts/post-code-test.sh:1053 — Tests reimplement the branch validation logic in a standalone function rather than exercising the production code. This matches existing test patterns in the file but means logic drift between the test helper and production code would not be caught.

Labels: PR fixes a bug in the code agent's branch validation logic (scripts/post-code.src.sh, skills/code-implementation/SKILL.md)

rh-hemartin · 2026-07-28T09:58:24Z

/fs-fix

fullsend-ai-coder · 2026-07-28T09:59:57Z

🤖 Finished Fix · ✅ Success · Started 9:59 AM UTC · Completed 10:05 AM UTC
Commit: c321286 · View workflow run →

fullsend-ai-coder · 2026-07-28T10:05:16Z

🔧 Fix agent — iteration 2 (human-triggered)

Addressed 2 of 5 review findings with code changes: rewrote contradictory documentation wording in SKILL.md and added a substring-matching edge-case test. Disagreed with 3 informational findings (protected-path requires human approval, authorization-change is intentional per #483, test-integrity refactor is out of scope).

Fixed (2):

documentation-coherence: contradictory wording about hardcoding main (skills/code-implementation/SKILL.md): Reworded the bolded guidance from 'Do not hardcode main as a fallback' to 'Do not skip discovery and assume main' to align with the ${DEFAULT_BRANCH:-main} fallback in the jq template two lines below.
edge-case: missing substring matching test (scripts/post-code-test.sh): Added substring-not-accepted test case verifying that agent_target='release' is rejected when allowed_list='release-1,release-2', confirming the comma-wrapping technique correctly rejects partial matches.

Disagreed (3):

test-integrity: tests reimplement logic rather than exercising production code: The reviewer notes this matches existing test patterns in the file. Changing the test approach to exercise production code directly would require a significant refactor of the test harness, which is out of scope for this PR.
protected-path: all changed files are under protected paths: Informational finding. Human approval is always required for protected-path changes — no code fix applies.
authorization-change: behavior changed from fail-closed to auto-correct: Informational finding. The behavior change is the intended fix per issue fix(code): defer target_branch until after default branch discovery #483. No code modification needed.

Tests: passed

_{Updated by fullsend fix agent}

- Reword documentation-coherence issue in SKILL.md: replace "Do not hardcode main as a fallback" with "Do not skip discovery and assume main" to align with the ${DEFAULT_BRANCH:-main} fallback in the jq template (documentation-coherence finding). - Add substring-not-accepted test case to verify comma-wrapping rejects partial matches like "release" when only "release-1" and "release-2" are in the allowed list (edge-case finding). Addresses review feedback on #488

rh-hemartin · 2026-07-28T10:07:45Z

/fs-fix rebase and resolve conflicts

fullsend-ai-coder · 2026-07-28T10:09:32Z

🤖 Finished Fix · ✅ Success · Started 10:09 AM UTC · Completed 10:14 AM UTC
Commit: b050b0f · View workflow run →

fullsend-ai-coder · 2026-07-28T10:14:07Z

🔧 Fix agent — iteration 3 (human-triggered)

Rebased agent/483-defer-target-branch onto origin/main, resolving one conflict in skills/code-implementation/SKILL.md where main renamed code-result.json to agent-result.json (PR #377). Resolution keeps our branch's ${DEFAULT_BRANCH:-main} variable substitution with the new filename. All 76 post-code tests pass. Bundle check passes.

Fixed (1):

rebase and resolve conflicts (skills/code-implementation/SKILL.md): Rebased branch onto origin/main (3 commits from PR fix(#376): standardize agent output filename to agent-result.json #377 for agent-result.json rename). Resolved merge conflict in skills/code-implementation/SKILL.md by keeping the branch's ${DEFAULT_BRANCH:-main} logic while adopting the new agent-result.json filename from main. All other files (post-code scripts) merged cleanly.

Tests: passed

_{Updated by fullsend fix agent}

Reorder branch discovery commands in SKILL.md so `gh repo view` is tried first, followed by `git rev-parse` and `git symbolic-ref` as fallbacks. Addresses review feedback on #488

- Reword documentation-coherence issue in SKILL.md: replace "Do not hardcode main as a fallback" with "Do not skip discovery and assume main" to align with the ${DEFAULT_BRANCH:-main} fallback in the jq template (documentation-coherence finding). - Add substring-not-accepted test case to verify comma-wrapping rejects partial matches like "release" when only "release-1" and "release-2" are in the allowed list (edge-case finding). Addresses review feedback on #488

rh-hemartin · 2026-07-30T09:51:27Z

/fs-fix rebase and resolve conflicts.

fullsend-ai-coder · 2026-07-30T09:53:17Z

🤖 Finished Fix · ❌ Failure · Started 9:53 AM UTC · Completed 10:01 AM UTC
Commit: bb81f63 · View workflow run →

The code agent hardcoded target_branch to "main" when branch discovery failed in the sandbox. The post-code script then rejected the output on repos with a non-main default branch (e.g. "master"). Two changes fix this: 1. SKILL.md step 3: Add fallback discovery commands (git symbolic-ref, gh repo view) so the agent tries multiple methods before writing target_branch. Document that the post-script auto-corrects when no explicit allowed list is configured. 2. post-code.src.sh: When CODE_ALLOWED_TARGET_BRANCHES is unset and the agent's target doesn't match the API-discovered default, auto- correct to the default branch with a warning instead of hard-failing. Hard failure is preserved when an explicit allowed list is set. Bundled script (post-code.sh) regenerated via make script-build. 8 new test cases cover the branch validation decision matrix. Note: pre-commit could not run in the sandbox (network restrictions). The post-script runs an authoritative pre-commit check on the runner. Closes #483

Reorder branch discovery commands in SKILL.md so `gh repo view` is tried first, followed by `git rev-parse` and `git symbolic-ref` as fallbacks. Addresses review feedback on #488

- Reword documentation-coherence issue in SKILL.md: replace "Do not hardcode main as a fallback" with "Do not skip discovery and assume main" to align with the ${DEFAULT_BRANCH:-main} fallback in the jq template (documentation-coherence finding). - Add substring-not-accepted test case to verify comma-wrapping rejects partial matches like "release" when only "release-1" and "release-2" are in the allowed list (edge-case finding). Addresses review feedback on #488

fullsend-ai-coder · 2026-07-30T10:01:06Z

⚠️ Post-fix script failed — Structured output processing failed (exit code 1)

The fix agent completed, but the post-fix script failed before finishing.

Workflow run: https://github.com/fullsend-ai/.fullsend/actions/runs/30532374812

Details:
process-fix-result.py failed with exit code 1 (bad input) for PR #488 in fullsend-ai/agents
Please check the workflow logs for full details and retry with /fs-fix if appropriate.

rh-hemartin · 2026-07-31T06:48:17Z

/fs-review

fullsend-ai-review · 2026-07-31T06:49:39Z

🤖 Finished Review · ✅ Success · Started 6:49 AM UTC · Completed 7:07 AM UTC
Commit: 5793196 · View workflow run →

fullsend-ai-review

See the review comment for full details.

fullsend-ai-retro · 2026-07-31T07:29:33Z

🤖 Finished Retro · ✅ Success · Started 7:29 AM UTC · Completed 7:40 AM UTC
Commit: 5793196 · View workflow run →

fullsend-ai-retro · 2026-07-31T07:40:55Z

Retro: PR #488 — fix(#483): defer target_branch until after default branch discovery

The workflow for issue #483 / PR #488 was largely successful. The code agent produced a solid 4-file fix with 8 test cases in ~8 minutes. The review agent found actionable issues (documentation-coherence, edge-case test coverage) that led to concrete code improvements via the fix agent. The human reviewer added value by requesting branch discovery command reordering (prioritizing gh repo view over git commands) and managing two rebases due to upstream changes.

Timeline summary: 8 agent runs (1 triage, 1 code, 4 fix, 2 review) over 3 days, with ~1 hour of active agent/human work on day 1. One fix run failed due to infrastructure (not agent error). PR merged after human approval.

Fix run 4 failure (infrastructure): The fix agent's rebase and push succeeded, but process-fix-result.py failed because it resolved the schema path (scripts/../schemas/fix-result.schema.json) relative to its own location in the content-addressed cache, where scripts/ and schemas/ live under different sha256 directories. This is already tracked by #551. Additionally, the failure comment didn't distinguish that the push had already succeeded — evidence supporting #534.

Review nondeterminism: The second review (on identical code for the relevant file) surfaced a new Medium finding ([behavior-description-incomplete] on docs/code.md:44) that the first review missed. This is the nondeterministic coverage pattern tracked by #511.

Documentation gap merged unaddressed: The review agent's second pass correctly identified that docs/code.md describes CODE_ALLOWED_TARGET_BRANCHES behavior as fail-on-mismatch, but the PR changed it to auto-correct-to-default. The human approved without addressing this finding — a conscious trade-off. The code agent not proactively updating this documentation is consistent with existing issues fullsend#4866 and fullsend#1733 about code agent documentation updates.

Branch discovery and sandbox compatibility: The human's first intervention was reordering branch discovery to prioritize gh repo view (API-based) over git rev-parse (git-based). This aligns with #489, which notes that git commands may not work under sandbox network policies.

No novel proposals. All identified improvement opportunities are already tracked by existing open issues. The workflow demonstrates that the triage → code → review → fix pipeline works well for this class of bug fix.

Reorder branch discovery commands in SKILL.md so `gh repo view` is tried first, followed by `git rev-parse` and `git symbolic-ref` as fallbacks. Addresses review feedback on #488

- Reword documentation-coherence issue in SKILL.md: replace "Do not hardcode main as a fallback" with "Do not skip discovery and assume main" to align with the ${DEFAULT_BRANCH:-main} fallback in the jq template (documentation-coherence finding). - Add substring-not-accepted test case to verify comma-wrapping rejects partial matches like "release" when only "release-1" and "release-2" are in the allowed list (edge-case finding). Addresses review feedback on #488

Reorder branch discovery commands in SKILL.md so `gh repo view` is tried first, followed by `git rev-parse` and `git symbolic-ref` as fallbacks. Addresses review feedback on fullsend-ai#488

- Reword documentation-coherence issue in SKILL.md: replace "Do not hardcode main as a fallback" with "Do not skip discovery and assume main" to align with the ${DEFAULT_BRANCH:-main} fallback in the jq template (documentation-coherence finding). - Add substring-not-accepted test case to verify comma-wrapping rejects partial matches like "release" when only "release-1" and "release-2" are in the allowed list (edge-case finding). Addresses review feedback on fullsend-ai#488

fullsend-ai-coder Bot requested a review from a team as a code owner July 28, 2026 09:16

fullsend-ai-coder Bot assigned rh-hemartin Jul 28, 2026

fullsend-ai-triage Bot mentioned this pull request Jul 28, 2026

fix(code-implementation): branch discovery and git fetch incompatible with sandbox network policy #489

Closed

This comment was marked as outdated.

Sign in to view

fullsend-ai-review Bot added requires-manual-review Review requires human judgment bug Something isn't working code-agent labels Jul 28, 2026

fullsend-ai-coder Bot force-pushed the agent/483-defer-target-branch branch from b050b0f to bb81f63 Compare July 28, 2026 10:14

fullsend-ai-coder Bot added 3 commits July 30, 2026 09:59

fix(skill): prioritize gh repo view for default branch discovery

1908dfd

Reorder branch discovery commands in SKILL.md so `gh repo view` is tried first, followed by `git rev-parse` and `git symbolic-ref` as fallbacks. Addresses review feedback on #488

fullsend-ai-coder Bot force-pushed the agent/483-defer-target-branch branch from bb81f63 to 5793196 Compare July 30, 2026 10:01

fullsend-ai-review Bot reviewed Jul 31, 2026

View reviewed changes

Comment thread scripts/post-code-test.sh

rh-hemartin approved these changes Jul 31, 2026

View reviewed changes

rh-hemartin added this pull request to the merge queue Jul 31, 2026

Merged via the queue into main with commit 95a6b43 Jul 31, 2026
14 checks passed

rh-hemartin deleted the agent/483-defer-target-branch branch July 31, 2026 07:27

fullsend-ai-coder Bot mentioned this pull request Jul 31, 2026

fix(#489): replace network-dependent git commands in skill #591

Merged

5 tasks

qodo-code-review Bot mentioned this pull request Aug 18, 2026

feat(risk): add PR risk assessment scoring to review pipeline #861

Merged

5 tasks

Uh oh!

Conversation

fullsend-ai-coder Bot commented Jul 28, 2026

Summary

Testing

Post-script verification

Uh oh!

rh-hemartin commented Jul 28, 2026

Uh oh!

fullsend-ai-coder Bot commented Jul 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

fullsend-ai-coder Bot commented Jul 28, 2026

🔧 Fix agent — iteration 1 (human-triggered)

Uh oh!

rh-hemartin commented Jul 28, 2026

Uh oh!

fullsend-ai-review Bot commented Jul 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

fullsend-ai-review Bot commented Jul 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Review

Findings

Medium

Low

Review

Findings

Medium

Low

Uh oh!

This comment was marked as outdated.

Uh oh!

rh-hemartin commented Jul 28, 2026

Uh oh!

fullsend-ai-coder Bot commented Jul 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

fullsend-ai-coder Bot commented Jul 28, 2026

🔧 Fix agent — iteration 2 (human-triggered)

Uh oh!

rh-hemartin commented Jul 28, 2026

Uh oh!

fullsend-ai-coder Bot commented Jul 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

fullsend-ai-coder Bot commented Jul 28, 2026

🔧 Fix agent — iteration 3 (human-triggered)

Uh oh!

rh-hemartin commented Jul 30, 2026

Uh oh!

fullsend-ai-coder Bot commented Jul 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

fullsend-ai-coder Bot commented Jul 30, 2026

Uh oh!

rh-hemartin commented Jul 31, 2026

Uh oh!

fullsend-ai-review Bot commented Jul 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

fullsend-ai-review Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

fullsend-ai-retro Bot commented Jul 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

fullsend-ai-retro Bot commented Jul 31, 2026

Retro: PR #488 — fix(#483): defer target_branch until after default branch discovery

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

fullsend-ai-coder Bot commented Jul 28, 2026 •

edited

Loading

fullsend-ai-review Bot commented Jul 28, 2026 •

edited

Loading

fullsend-ai-review Bot commented Jul 28, 2026 •

edited

Loading

fullsend-ai-coder Bot commented Jul 28, 2026 •

edited

Loading

fullsend-ai-coder Bot commented Jul 28, 2026 •

edited

Loading

fullsend-ai-coder Bot commented Jul 30, 2026 •

edited

Loading

fullsend-ai-review Bot commented Jul 31, 2026 •

edited

Loading

fullsend-ai-retro Bot commented Jul 31, 2026 •

edited

Loading