Skip to content

Code agent missing fullsend-check-output self-validation before exit #2777

Description

@rh-hemartin

What happens

The code agent (skills/code-implementation/SKILL.md) writes its structured output to $FULLSEND_OUTPUT_DIR/$FULLSEND_OUTPUT_FILE but never runs fullsend-check-output to validate it against the schema before exiting.

All other agents (triage, review, fix, prioritize, retro) include a self-check instruction block after writing their output JSON. The code agent is the only one missing it.

Local transcript analysis of 6 runs across agent types confirmed that agents only execute the self-check when their definition explicitly instructs them to.

What should happen

skills/code-implementation/SKILL.md should include the standard self-check block after the output write section (around line 201), consistent with the other agents:

fullsend-check-output "${FULLSEND_OUTPUT_DIR}/${FULLSEND_OUTPUT_FILE}"

With the same retry guidance: read the error output, fix the JSON, re-run. Exit with best-effort JSON after 3 failed attempts.

Context

The code agent's harness (harness/code.yaml) has no validation_loop — so there's no host-side retry either. A malformed code-result.json goes straight to the post-script without any schema check.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    Status
    Todo

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions