Skip to content

Comments

Fix new_job workflow issues and update learning-agents skills#223

Merged
nhorton merged 6 commits intomainfrom
learning_agent_fixes
Feb 20, 2026
Merged

Fix new_job workflow issues and update learning-agents skills#223
nhorton merged 6 commits intomainfrom
learning_agent_fixes

Conversation

@nhorton
Copy link
Contributor

@nhorton nhorton commented Feb 18, 2026

Summary

  • test step: Added instructions to write test feedback to .deepwork/tmp/test_feedback.md — previously the step never told the agent to create this declared output, so finished_step would fail
  • test reviews: Replaced conversational review criteria (e.g. "User Satisfied") with file-based criteria that actually review the feedback file
  • make_new_job.sh: Removed stale "Run deepwork install" from generated AGENTS.md (CLI command was removed)
  • iterate + learn steps: Added script creation/fix step so the scripts output is actually addressed in instructions
  • learning-agents skills: Removed deprecated disable-model-invocation and allowed-tools frontmatter fields; added restart warning to create-agent

Test plan

  • Run deepwork_jobs/new_job workflow through all 4 steps and verify test step writes feedback file
  • Verify finished_step accepts the test feedback file output
  • Run make_new_job.sh and check generated AGENTS.md has no install reference
  • Verify learning-agents skills still load correctly after frontmatter changes

🤖 Generated with Claude Code

nhorton and others added 6 commits February 18, 2026 13:58
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
New agent for reviewing PRs against DeepWork's architectural patterns,
naming conventions, and process standards. Seeded with topics covering
MCP workflow patterns and the quality review system.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- test.md: Add Step 5 to write feedback to .deepwork/tmp/test_feedback.md
- job.yml: Replace conversational test review criteria with file-based ones
- make_new_job.sh: Remove stale "deepwork install" reference from AGENTS.md
- iterate.md/learn.md: Add script creation/fix step for automation opportunities
- learning-agents skills: Remove deprecated frontmatter fields, add restart warning to create-agent

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The learn skill was using general-purpose subagent_type instead of the
dedicated learning-agents:learning-agent-expert agent, which has the
domain knowledge needed for issue identification and investigation.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- post_task.sh: Use correct hookSpecificOutput structure for additionalContext
  so the reminder actually surfaces to the calling agent
- post_task.sh: Normalize agent name (lowercase + hyphens) to match directory
  naming when frontmatter name differs (e.g. "Consistency Reviewer")
- learn SKILL.md: Use learning-agents:learning-agent-expert subagent_type
  instead of general-purpose

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
@nhorton nhorton merged commit 8367e97 into main Feb 20, 2026
4 checks passed
@nhorton nhorton deleted the learning_agent_fixes branch February 20, 2026 20:46
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant