CommunityResearch & Data Analysisgithub.com

DevArtsLab/devartslab-linear-research-refine

Run one bounded refinement pass on a living Linear research ticket. Use for tickets structured with findings, open questions, a decision log, and a refinement protocol (e.g., DEV-47). Verify decaying claims against current sources, update findings in place, and log resolved decisions with evidence.

What is devartslab-linear-research-refine?

devartslab-linear-research-refine is a Claude Code agent skill that run one bounded refinement pass on a living Linear research ticket. Use for tickets structured with findings, open questions, a decision log, and a refinement protocol (e.g., DEV-47). Verify decaying claims against current sources, update findings in place, and log resolved decisions with evidence.

Works with~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/DevArtsLab/agent-skills/tree/main/skills/devartslab-linear-research-refine

Installed? Explore more Research & Data Analysis skills: affaan-m/uspto-database, affaan-m/scholar-evaluation, affaan-m/literature-review · View all 6 →

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

Linear Research Refinement

Runs a single daily-quality research pass against a living Linear research ticket. Default target: DEV-47 (productize job tracker as multi-tenant SaaS on Cloudflare, DevArts Lab workspace). Pass a different issue identifier to reuse for another ticket.

Pass protocol

  1. Fetch the ticket via linear-mcp-server (get_issue or list_issues by identifier). Read the full description and all comments first.

  2. Choose one focus area: the highest-priority open question with low confidence, or the oldest unverified claim in the findings. Go deep on one area rather than touching everything.

  3. Research with authoritative sources only: official docs, changelogs, vendor pricing pages, upstream repos/issues. Use web_search + webfetch. Record each source URL and date checked.

  4. Update the ticket in place (save_issue patch):

    • Mark verified claims with [verified YYYY-MM-DD: source]
    • Correct stale or wrong claims
    • Move answered questions into the Decision Log with: date, decision, evidence, confidence, and what would change the decision
    • Add new open questions discovered (why it matters + priority)
    • Update phasing/plans if evidence warrants it
  5. One comment per pass (save_comment): focus area, what changed, what to look at next. Flag product-judgment calls (pricing, scope, privacy trade-offs) for the user instead of deciding them.

  6. Stay bounded: a focused increment. Daily passes compound.

Rules

  • Research and ticket updates only. Do not modify code or repos.
  • Never invent sources or capabilities. Mark unverifiable claims as unverified with the missing evidence.
  • Preserve the ticket's structure; edit in place, never rewrite wholesale.
  • If every question is resolved: re-verify the oldest verified claims (they decay) or deepen the implementation plan.
  • The workspace repo may be read for context but not edited.

Scheduling

Scheduled copies are generated per-project by the agent-research-kit (~/projects/devartslab/agent-research-kit/install.sh), which renders templates/prompt.md.tmpl with each project's issue ID. This file is the interactive reference; the kit's prompt template is the scheduled master.

Individual skills in this repo

This repo contains 2 individual skills — each has its own dedicated page.

Related Skills