docs(#484): document CODE_ALLOWED_TARGET_BRANCHES in AGENTS.md

[fullsend-ai-coder]

Summary

• Add a "Code agent configuration" section (§7) to AGENTS.md documenting the CODE_ALLOWED_TARGET_BRANCHES environment variable
• Document the variable's format (comma-separated branch names or * wildcard), default behavior (auto-detected via GitHub API, falls back to main), and where to set it (runner environment, passed through runner_env in harness/code.yaml)
• Renumber the existing "Skill resolution" section from §7 to §8

Context

CODE_ALLOWED_TARGET_BRANCHES controls which branches the code agent is allowed to push to. It is defined in harness/code.yaml:52 and validated in scripts/post-code.src.sh, but was not documented anywhere user-facing. All documented details were verified against the actual source code.

Testing

• Documentation-only change — no behavioral changes
• Verified documented format, default, and behavior match the implementation in scripts/post-code.src.sh
• Secret scan passed
• Gitlint passed

---

Closes #484

Post-script verification

• Branch is not main/master (agent/484-document-allowed-branches)
• 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]

/fs-fix the changes have to go under docs/code.md in the variables section.

[fullsend-ai-coder]

🤖 Finished Fix · ✅ Success · Started 9:28 AM UTC · Completed 9:32 AM UTC
Commit: fae08a0 · View workflow run →

[fullsend-ai-coder]

🔧 Fix agent — iteration 1 (human-triggered)

Moved CODE_ALLOWED_TARGET_BRANCHES documentation from AGENTS.md to docs/code.md under the Variables section per human instruction. Reverted AGENTS.md to its original state with correct section numbering.

Fixed (2):

1. move documentation to docs/code.md variables section (docs/code.md): Moved CODE_ALLOWED_TARGET_BRANCHES documentation from AGENTS.md section 7 to docs/code.md under the Variables section, as requested by reviewer
2. revert AGENTS.md changes (AGENTS.md): Removed the Code agent configuration section from AGENTS.md and restored section numbering (Skill resolution back to section 7)

Tests: passed

_{Updated by fullsend fix agent}

[rh-hemartin]

/fs-fix Use a table as the other docs, also the env var should be passed to env.runner not runner_env, the later is deprecated.

[fullsend-ai-coder]

🤖 Finished Fix · ✅ Success · Started 9:37 AM UTC · Completed 9:42 AM UTC
Commit: 2f054fd · View workflow run →

[fullsend-ai-coder]

🔧 Fix agent — iteration 2 (human-triggered)

Addressed both human review findings: reformatted the Variables section to use a standard horizontal table (matching review.md) and replaced the deprecated runner_env reference with env.runner.

Fixed (2):

1. Use a table format matching other docs (e.g. review.md) (docs/code.md): Replaced the vertical Aspect/Detail table and heading with a horizontal Variable/Description/Default/Valid-values table matching the pattern used in review.md
2. Use env.runner instead of deprecated runner_env (docs/code.md): Changed the reference from 'runner_env' to 'env.runner' in the Where-to-set description, matching the newer harness format used in fix.yaml and scribe.yaml

Tests: passed

Decision points

• Condensed the prose description and rejection-behavior paragraph into the table's Description cell to keep the section compact (alternatives: Keep separate prose paragraphs below the table; rationale: review.md uses a single table row with inline description and no extra paragraphs — matching that pattern exactly)

_{Updated by fullsend fix agent}

[rh-hemartin]

/fs-review

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 9:46 AM UTC · Completed 9:58 AM UTC
Commit: 917e643 · View workflow run →

[fullsend-ai-review]

Review

Findings

Medium

• [technical documentation accuracy] docs/code.md:42 — The documentation states the variable is "Set via env.runner in harness/code.yaml" but the actual YAML key in harness/code.yaml (line 51) is runner_env, not env.runner. A user following this documentation to configure the variable would look for the wrong key in the YAML file.
  Remediation: Change env.runner to runner_env in the description: "Set via runner_env in harness/code.yaml (never injected into the sandbox)."

Low

• [scope-mismatch-metadata] — PR title and body claim to modify AGENTS.md (specifically "Add a Code agent configuration section (§7) to AGENTS.md" and "Renumber the existing Skill resolution section from §7 to §8"), but the actual change is to docs/code.md. The choice of docs/code.md is appropriate per the issue's flexibility clause, but the PR metadata misrepresents what was actually implemented.
  Remediation: Update PR title and body to accurately reflect that the change was made to docs/code.md, not AGENTS.md. Remove the claim about renumbering sections.

---

Labels: PR adds documentation for the code agent's CODE_ALLOWED_TARGET_BRANCHES configuration variable

[fullsend-ai-review]

See the review comment for full details.

[rh-hemartin]

/fs-fix

[fullsend-ai-coder]

🤖 Finished Fix · ✅ Success · Started 10:03 AM UTC · Completed 10:09 AM UTC
Commit: 917e643 · View workflow run →

[fullsend-ai-coder]

🔧 Fix agent — iteration 3 (human-triggered)

Fixed 1 of 2 review findings. Changed env.runner to runner_env in docs/code.md to match the actual harness YAML key. Disagreed with the PR metadata finding as it requests non-code changes outside the fix agent's scope.

Fixed (1):

1. technical documentation accuracy: env.runner vs runner_env (docs/code.md): Changed env.runner to runner_env in the CODE_ALLOWED_TARGET_BRANCHES documentation. The actual YAML key in harness/code.yaml (line 51) is runner_env, and all harness files in the codebase use runner_env consistently. The documentation now accurately reflects the codebase.

Disagreed (1):

1. scope-mismatch-metadata: PR title and body reference AGENTS.md: This finding requests changes to PR title and body (metadata), not to source code. The fix agent's scope is limited to code changes; PR metadata updates should be handled by a human or the post-script. The code change in docs/code.md is correctly placed.

Tests: passed

_{Updated by fullsend fix agent}

[fullsend-ai-retro]

🤖 Finished Retro · ✅ Success · Started 10:21 AM UTC · Completed 10:39 AM UTC
Commit: a9b1e60 · View workflow run →

[fullsend-ai-retro]

Retro: PR #487 — Document CODE_ALLOWED_TARGET_BRANCHES

A 3-line documentation change to docs/code.md required 3 fix iterations before merging. The rework was driven by two factors: (1) the issue directed documentation to the wrong file (AGENTS.md instead of docs/code.md), and (2) the human reviewer gave incorrect guidance about a deprecated config key name (runner_env vs env.runner) that contradicted the actual codebase.

What went well

• Review agent caught a human-introduced error. The human reviewer instructed the fix agent to change runner_env to env.runner, citing a deprecation (ADR 0055). The review agent correctly identified that the actual YAML in harness/code.yaml still uses runner_env, flagging the discrepancy as a medium-severity finding. The fix agent then correctly reverted the change. This is additional evidence for autonomy-readiness tracking in Track autonomy readiness: review agent outperforms human review on mechanical consistency checks in konflux-ci/architecture PR #368 fullsend#4995 and Track autonomy evidence: review agent competently handles docs-only PRs #427 — the review agent outperformed human review on a factual accuracy check.
• Fix agent performed well across all 3 iterations — correctly interpreting human instructions, relocating content, reformatting, and even correctly disagreeing with a low-severity metadata finding as out-of-scope.

Evidence for existing issues (not filed as new proposals)

• agents#436 (code agent cross-referencing existing content): The code agent read harness/code.yaml (which has doc: docs/code.md) but did not use the doc: field to infer the correct documentation file. While Code agent should cross-reference existing content when adding documentation sections #436 addresses content quality within a file, the file-selection gap is a related but distinct issue that I'm proposing below.
• agents#452 (triage enumerating doc touchpoints): The triage agent echoed the issue's incorrect file target (AGENTS.md) without correction. Triage agent should enumerate documentation touchpoints for CLI changes #452 is scoped to CLI changes specifically; the pattern here is broader.
• Track autonomy readiness: review agent outperforms human review on mechanical consistency checks in konflux-ci/architecture PR #368 fullsend#4995, agents#427: The review agent catching the human-introduced env.runner error is further evidence that the review agent handles factual accuracy checks well on documentation PRs.

Proposals filed

• Code agent should check harness doc: field to determine documentation placement (in fullsend-ai/agents)
• Fix agent should support updating PR metadata when scope changes (in fullsend-ai/agents)