@zack/devwork-copilot

Interactive GitHub maintainer copilot with explicit approval gates for write actions, packaged maintainer-review orchestration, workflow continuity, and a durable maintainer behavior profile.

Install
agentpm install @zack/devwork-copilot@0.1.7
Load
import { loadAgent } from '@agentpm/sdk';
const agent = await loadAgent('@zack/devwork-copilot@0.1.7');
Installs
15
All time
Stars
1
Unstarred
Latest stable
Today
v0.1.7

devwork-copilot

devwork-copilot is a published AgentPM agent package for maintainer and repository review workflows in Python.

Version 0.1.7 adds the exact direct Tool dependency @zack/summarize-text@0.1.9, which uses AgentPM-managed Python dependencies instead of a bundled environment. Use AgentPM CLI 0.1.37 or newer and a Python 3.11+ interpreter for this version.

It includes a Loop for:

  • inspect -> draft -> resolve/handoff maintainer orchestration with an explicit revise path back into drafting

It composes skills for:

  • issue triage workflow

It includes Knowledge for:

  • maintainer guidance retrieval across triage, labeling, and comment-drafting tasks

It includes Memory for:

  • maintainer workflow continuity across profile preferences, active work threads, and follow-up notes

It includes a Profile for:

  • concise maintainer tone, explicit approval boundaries, and pragmatic follow-through guidance

It includes authored bindings for:

  • a shared maintainer behavior profile
  • maintainer-state cleanup as a global Memory capability
  • phase-scoped Tool, Knowledge, Skill, and Memory surfaces
  • a maintainer-specific consumer-context filename

The direct summarizer is available during inspect-issue and draft-comment. It can condense supplied issue text or maintainer notes into a requested word budget. The final resolve-or-handoff phase keeps Tool access disabled.

GitHub capability enters through the packaged issue-triage-playbook Skill, which depends on the published @zack/github-issues Tool.

Example prompts:

  • Review the latest GitHub issues for a repository, group them by theme, and summarize the highest-priority work.
  • Read a set of repository issues and produce a concise maintainer briefing with action items and open questions.
  • Review an issue thread, retrieve the most relevant maintainer guidance, and draft a short comment that sets expectations clearly.
  • Review the current repository issues and summarize what the next maintainer should know, including current priorities, open follow-ups, unresolved questions, and anything that should carry forward.
  • Inspect the issue thread, decide whether the next step is to draft a maintainer comment or hand the work off, and explain which packaged loop phase that maps to.
  • Review the issue thread, summarize the current maintainer position, and draft the next follow-up comment.

Local development

The source code for this agent package can be found here

Prepare and publish

Publish @zack/summarize-text@0.1.9 first, using its reviewed author lock, before publishing this Agent as @zack/devwork-copilot@0.1.7. From this directory:

agentpm lint --strict
agentpm publish --dry-run
agentpm publish --namespace zack

Publishing the Agent packages its composition; it does not install its Tools or create a consumer lock. Keep the existing published Agent versions unchanged.

Install and verify

After both package versions are published, run these commands from a consumer workspace. To update the checked-in Harness example, use app-harness-minimal-agent/ as that workspace:

agentpm install @zack/devwork-copilot@0.1.7
export OPENAI_API_KEY='your-key'
agentpm run @zack/summarize-text@0.1.9 --input '{"text":"The maintainer reviewed the bug report, reproduced the problem, and requested a regression test before merging the fix.","max_words":60}'
agentpm harness --json
mkdir -p reports
agentpm harness --headless --input 'During inspection, call summarize-text on this supplied note with a 60-word budget: The maintainer reproduced the bug and requested a regression test before merging. Hand off when the summary is complete.' --report reports/summarizer-check.json

Direct execution and Harness model/Tool requests use the OpenAI API and can incur charges. GITHUB_TOKEN is needed only for live GitHub actions; this supplied-text check does not require fetching an issue. Confirm the report/trace contains an actual summarize-text Tool call rather than only a model-written summary. The supplied-text check may hand off after inspection without entering later phases, which can require additional configured capabilities.

Installation updates the workspace's dependency request/portable agent.lock and provisions the summarizer's hash-verified Python environment under .agentpm/runtimes/python/. Review and commit that generated consumer state after a successful real Registry install. Keep consumer locks/installed package digests from previously published versions intact until the new versions exist; do not replace their version strings or invent integrity values by hand.