joenandez/spectre/spectre-create-pr

Claude Code and agent skill by joenandez/spectre

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Generate a grounded draft pull request from the actual diff and open it via gh. Use when wrapping up a branch or writing a PR description; not for Ship-owned cleanup, rebase, or delivery sequencing.

Install

npx vskill@latest install joenandez/spectre/spectre-create-pr

SKILL.md

create_pr

Produce a grounded draft PR.

Inputs

  • $ARGUMENTS: TARGET_BRANCH (default origin/main), feedback focus, verification, all-or-none EXPECTED_BASE_SHA/EXPECTED_HEAD_SHA/EXPECTED_DIFF_SHA256, or --orchestrated.
  • Orchestrated --pr-phase pending|final-update: pending needs complete candidate tuple, local verification RUNNING, and returns URL/body; final-update needs its URL/body, same tuple, and FINAL_VERIFICATION_SUMMARY, updating existing draft Testing only. A parent may supply a frozen work-ID set. Orchestrated Create PR does not write a work record; an unchanged candidate is a no-op.
  • Resolve just-in-time: branch (not main/master), fetch target and derive PR_BASE, BASE_SHA, HEAD_SHA, canonical git-diff-v1 DIFF_SHA256 over binary/full-index/no-color/no-renames {BASE_SHA}...{HEAD_SHA}, commits, branch/commit issue ref, gh, unpushed commits.
  • Expected fields are all-or-none. After fetch compare live tuple and clean candidate worktree (committed review/proof artifacts allowed; ignored lifecycle excluded). Tracked or non-ignored untracked changes return PR_CANDIDATE_STALE before push, create, or edit. Without gh, return title/body for manual draft.

Working Set

Target-to-HEAD diff and commit log, issue reference, and any GitHub PR template (required sections win).

Outputs + DONE

  • Conventional type(scope): summary: <70 chars, imperative, lowercase after colon, no period; grounded body scales to change. Standalone/pending opens gh pr create --draft only without an open branch PR, else updates it.

DONE: fetched tuple is verified; every factual claim is grounded; Testing honestly reflects diff tests and verification, never turns advisory non-green into pass; sourced/placeholder Why; no secret/credential/PII; only a draft is opened or updated, never duplicated.

Method / guardrails

  1. Ground: What is behavior; Why is issue, commits, branch, or <!-- WHY: motivation not found in commits/issue — fill in -->; visible How/trade-offs; Testing reports supplied verification, changed tests, or none.
  2. Scale: trivial What/Why/Closes; standard Summary/Changes/Testing/Closes; complex adds trade-offs, breaking/rollback, UI/CLI evidence, reviewer focus. Derive type/scope from change; add found issue links only.
  3. Verify before side effects: map every claim to diff/commit/issue, drop unsupported claims/secrets, verify tuple, clean candidate, and this branch's own open PR (gh pr list --head; never another branch's or closed), then push.
  4. Work-record ownership: in orchestrated mode return PR evidence to Ship without writing a work record. In standalone mode only, when no Execute or Ship parent owns history, invoke Skill(spectre-work-record) once at the terminal boundary after the draft exists; that write and explicit historical correction are unchanged; report capture failure/recovery input without blocking draft. Refresh a stale parent candidate; a draft is not merged.
  5. Draft lifecycle: pending grounds RUNNING, pushes, updates that open PR, else creates the draft, and returns its PR identity/URL against the frozen work-ID set for Ship to associate. Final-update rechecks its tuple/clean candidate; if repairs changed the tuple, refresh candidate-sensitive claims under freshness, grounding, secret gates, verify clean repaired HEAD, pushes, re-resolves/rechecks live tuple, then gh pr edit only Testing from FINAL_VERIFICATION_SUMMARY; never mark ready.

Handoff

Return URL/body or PR_CANDIDATE_STALE; --orchestrated: no user step.

HandoffDetails
🧭 Current phaseDone
📦 What was just doneResult
▶️ Proposed next stepRender resolved action.

Standalone: review the PR.

Escalate-If

  • Branch is main/master or has no target-ahead commits; candidate tuple is incomplete/different; branch/target/remote is unsafe; or the diff has secrets/PII.
  • Large intent is ungrounded: use the Why placeholder and draft, never fabricate.

Security

  • Security scanPassed, 90 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

Repository stars
164
Repository forks
16
7-day trend score
31

Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/joenandez/spectre/spectre-create-pr/badge)](https://verified-skill.com/skills/joenandez/spectre/spectre-create-pr)

Something wrong with this skill or its page? Tell us.