fix(#22): enforce structured output in code agent

The code agent was missing structured output enforcement, causing
validation_loop failures. The review and fix agents both have explicit
structured output sections and mandatory final validation steps, but
the code agent had neither.

Changes:
- agents/code.md: add "Structured output" section documenting the
  requirement to produce code-result.json and validate it with
  fullsend-check-output before exiting. Update failure handling to
  reference structured output in the handoff contract.
- skills/code-implementation/SKILL.md: add step 11 as a mandatory
  final step to validate structured output against the schema. Add
  forward reference from step 3 (where the file is first written)
  to step 11. Update progress markers list. Require structured
  output even for partial work exits.

Closes #22

Assisted-by: Claude
Signed-off-by: Wayne Sun <gsun@redhat.com>

Author: waynesun09

agents/code.md

@@ -92,17 +92,38 @@ the review agent — if the triage was wrong, your code will fail review.
 - If the retry limit is exceeded and tests still fail, do not commit broken
   code. Stop. The post-script reports the failure.
 
+## Structured output
+
+You MUST produce a JSON file at `$FULLSEND_OUTPUT_DIR/$FULLSEND_OUTPUT_FILE`
+(typically `code-result.json`) that documents the target branch for PR
+creation. The `code-implementation` skill describes the schema and the exact
+step where you write it. The post-script reads this file to determine which
+branch to target the PR against. Without this file, the validation loop
+rejects the run and retries.
+
+After writing the file, validate it before exiting:
+
+```bash
+fullsend-check-output "${FULLSEND_OUTPUT_DIR}/${FULLSEND_OUTPUT_FILE}"
+```
+
+If validation fails, read the error output, fix the JSON file, and
+re-run the check. If it still fails after 3 attempts, write the best
+JSON you have and exit.
+
 ## Failure handling
 
 Secret scanning is **non-negotiable**. The `scan-secrets` helper runs before
 tests on every verification pass. If secrets are detected — or if the helper
 script is missing — hard stop. Do not improvise a replacement or skip the scan.
 
 Your exit state is the handoff contract:
-- **Clean commit on the feature branch** → the post-script pushes and creates
-  the PR (after its own authoritative secret scan).
+- **Clean commit on the feature branch + valid structured output** → the
+  post-script pushes and creates the PR (after its own authoritative secret
+  scan).
 - **No commit** → the post-script reads your transcript and exit code to
-  report the failure.
+  report the failure. Structured output should still be written when possible
+  so the post-script knows which branch was targeted.
 
 ## Detailed implementation procedure
 

skills/code-implementation/SKILL.md

@@ -56,7 +56,7 @@ echo "::notice::STEP <N>: <title>"
 ```
 
 This uses GitHub Actions annotation syntax so it surfaces in the run
-summary. **Do this at steps 1, 3, 5, 9a, 9b, 9c, and 10.**
+summary. **Do this at steps 1, 3, 5, 9a, 9b, 9c, 10, and 11.**
 
 ## Time budget
 
@@ -186,8 +186,8 @@ use that branch. Otherwise, determine the repo's default branch:
 git rev-parse --abbrev-ref origin/HEAD | cut -d/ -f2
 ```
 
-Write your chosen branch to the structured output file so the post-script
-knows which branch to target the PR against:
+Write your chosen branch to the structured output file **early** so
+it is available even if the agent hits a timeout or error later:
 
 ```bash
 mkdir -p "${FULLSEND_OUTPUT_DIR}"
@@ -198,9 +198,8 @@ cat > "${FULLSEND_OUTPUT_DIR}/${FULLSEND_OUTPUT_FILE}" <<RESULT
 RESULT
 ```
 
-Write this output early (during planning, after determining the target
-branch) so it is available even if the agent hits a timeout or error later.
 The post-script validates this against the repo's allowed branches.
+You will re-validate this file in the final step (step 11).
 
 ### 4. Check for existing branch
 
@@ -739,6 +738,43 @@ message. The post-script runs an authoritative pre-commit on the runner.
 **Do not push the branch.** The post-script handles pushing, PR creation,
 and failure reporting.
 
+### 11. Validate structured output
+
+**This step is MANDATORY.** The harness runs a validation loop that
+checks `$FULLSEND_OUTPUT_DIR/$FULLSEND_OUTPUT_FILE` against
+`schemas/code-result.schema.json`. If validation fails, the harness
+retries the agent. Producing a valid output file is not optional.
+
+You wrote the initial output file in step 3. Confirm it still exists
+and contains the correct target branch:
+
+```bash
+echo "::notice::STEP 11: Validate structured output"
+cat "${FULLSEND_OUTPUT_DIR}/${FULLSEND_OUTPUT_FILE}"
+```
+
+The file must be valid JSON with exactly one field:
+
+```json
+{
+  "target_branch": "main"
+}
+```
+
+**Schema compliance:** The schema uses `additionalProperties: false`.
+Only the `target_branch` field is allowed. Any extra fields will cause
+validation to fail.
+
+Validate the output against the schema:
+
+```bash
+fullsend-check-output "${FULLSEND_OUTPUT_DIR}/${FULLSEND_OUTPUT_FILE}"
+```
+
+If validation fails, read the error output, fix the JSON file, and
+re-run the check. If it still fails after 3 attempts, write the best
+JSON you have and exit.
+
 ## Partial work
 
 If you hit a token limit or context window boundary before completing the
@@ -748,6 +784,10 @@ code is caught at the review stage, not the implementation stage. The commit
 message should note that the work is partial (e.g., "partial implementation"
 in the description) so the review agent and post-script can act accordingly.
 
+**Structured output is still required for partial work.** The output file
+written in step 3 must exist and be valid. Run `fullsend-check-output`
+(step 11) even when exiting early.
+
 ## Constraints
 
 The agent definition (`agents/code.md`) is the authoritative list of