Github-research-plugin document-findings
Write GitHub research findings into a planning/ folder in the current repository, always date-stamped so results stay interpretable as the tooling landscape evolves. Creates planning/ if it does not already exist. Records the exact gh queries used so the research is reproducible.
git clone https://github.com/danielrosehill/github-research-plugin
T=$(mktemp -d) && git clone --depth=1 https://github.com/danielrosehill/github-research-plugin "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/document-findings" ~/.claude/skills/danielrosehill-github-research-plugin-document-findings && rm -rf "$T"
skills/document-findings/SKILL.mdDocument GitHub Research Findings
The tooling landscape on GitHub evolves quickly — a top project today may be abandoned in six months, and new entrants appear constantly. Any research write-up must be date-stamped so future readers (including future-you) know the context in which the recommendation was made.
When to use
Trigger phrases include:
- "Save these findings"
- "Document this research"
- "Write this to the planning folder"
- "Add a note in the repo about this research"
Procedure
1. Locate or create the planning folder
In the current repository (working directory), look for one of:
planning/docs/planning/.planning/
If none exist, create
planning/ at the repo root.
Inside the planning folder, create or reuse a subfolder:
planning/github-research/
This keeps research artefacts grouped and out of the way of general planning docs.
2. Name the file with the date
File naming convention:
planning/github-research/YYYY-MM-DD-<short-topic-slug>.md
Examples:
planning/github-research/2026-04-16-mcp-server-frameworks.mdplanning/github-research/2026-04-16-markdown-to-pdf-cli.md
Use
date +%Y-%m-%d to get today's date — never hard-code a year.
If a file for the same topic already exists on the same date, append a short suffix (e.g.
-v2) rather than overwriting.
3. Write the document
Use this template:
# GitHub Research: <Topic> **Research date:** YYYY-MM-DD **Researcher:** Claude Code (github-research plugin) **Goal:** <one-sentence statement of what the user was trying to find / decide> > ⚠️ Note on freshness: GitHub's tooling landscape evolves quickly. Findings below reflect repository state on the research date. Re-run the queries before acting on recommendations more than ~3 months old, especially for AI/ML projects. ## Queries run Exact `gh` / `gh api` commands used, so the research is reproducible: ```bash <paste each command>
Top candidates
| Repo | Stars | Last push | Open issues | License | Language | Notes |
|---|---|---|---|---|---|---|
| owner/repo | 1.2k | 3 days ago | 14 | MIT | Python | Active, strong CI |
| ... | ... | ... | ... | ... | ... | ... |
Detailed evaluation
owner/repo
- Verdict: recommend / maybe / avoid
- Strengths: ...
- Concerns: ...
- Fit for our use case: ...
(Repeat per candidate)
Recommendation
<Concrete, opinionated recommendation. Which to try first, or "none are a good fit — build from scratch, here's why".>
Open questions / follow-ups
- <things the user should check before committing>
### 4. Commit hygiene (do not auto-commit) Write the file only. Do **not** run `git add` / `git commit` unless the user explicitly asked. Let the user review the document first. After writing, report the path of the created file so the user can open it. ### 5. Link back from the repo's CLAUDE.md (optional) If the repo has a `CLAUDE.md` and the user indicates this research will inform ongoing work, offer (don't auto-apply) to add a line under a "Prior research" section: ```markdown ## Prior research - [YYYY-MM-DD: <Topic>](planning/github-research/YYYY-MM-DD-<slug>.md)
Guidance
- Always include the date in both the filename and the document body. The whole point of this skill is reproducibility across time.
- Always include the exact queries run. Without them the findings cannot be re-verified when they age.
- Be honest about limits. If the search was narrow, say so. If results were sparse, say so. Do not invent projects to pad the list.
- Prefer opinionated recommendations over neutral catalogues. The user asked for research to make a decision, not a Wikipedia article.