Agent-Skills.md

Agent Skills: write-gh-issues-prs

[DEPRECATED] This skill is no longer used. Feature tracking has migrated to PROJECT.md.

UncategorizedID: JoernStoehler/ai-forecasting-hackathon/write-gh-issues-prs

Install this agent skill to your local

pnpm dlx add-skill https://github.com/JoernStoehler/ai-forecasting-hackathon/tree/HEAD/.claude/skills/write-gh-issues-prs

Skill Files

Browse the full folder contents for write-gh-issues-prs.

Download Skill

Loading file tree…

.claude/skills/write-gh-issues-prs/SKILL.md

Skill Metadata

Name
write-gh-issues-prs
Description
"[DEPRECATED] This skill is no longer used. Feature tracking has migrated to PROJECT.md."

DEPRECATED: GitHub Issues Migration

This skill is deprecated as of 2026-01-18.

We have migrated from GitHub Issues & Milestones to PROJECT.md-based feature tracking.

Why the change?

  • Agents can read PROJECT.md directly (no API keys needed)
  • Single source of truth lives with the code
  • Atomic updates (code + status in same PR)
  • Version controlled feature evolution
  • Lower overhead for agent-driven development

New workflow:

  1. Read PROJECT.md to understand feature status (🟢🟡🔵⚪🔴)
  2. Implement feature
  3. Update PROJECT.md in same PR
  4. Commit: "feat: implement X, mark complete in PROJECT.md"

For PR descriptions:

See .github/PULL_REQUEST_TEMPLATE.md (still relevant for PR content).

Old Documentation (archived)

~~We use GitHub issues and PRs to coordinate work between agents and the project owner.~~

  • Best practices with regards to content are explained in the templates: .github/ISSUE_TEMPLATE/task.md, .github/PULL_REQUEST_TEMPLATE.md.
  • Use GitHub CLI with --body-file to avoid shell quoting pain.
  • Agents start with zero context about the project before onboarding, and they will focus on onboarding and reference materials relevant to their task. We have to mention required readings explicitly in the github issues so that agents pick them up.
  • Style:
    • Literal correctness, unambiguity, clarity are important. We don't have much interactiveness between author and worker agents, so misunderstandings must be avoided up front.
    • Calibrated confidence: predictions have relative strengths (probabilities, expected utility), and ideas are predictions about what subgoals, metrics, constraints, or even concrete steps will help the overall thesis project how much. We don't expunge the uncertainty and relative strength, but clearly communicate what we believe confidently is a must-have, and what is a speculative idea, or even just a first guess to try and throw out if unsuitable.
    • Most PRs are reviewed by the project owner only. We help the project owner by offering skimmable extra commentary and not leaving out insights we have that could cut down the time the project owner has to spend understanding, evaluating, or handing back the PR.
    • We use numbered/lettered lists to make references easier, since line numbers are not quite as stable.
  • GitHub Identity: All agents share the GitHub identity of the project owner, thus various fields in PRs and issues lose their meaning (e.g. author, assignee). Instead we use footers and body text to clarify these aspects where relevant.