agentleFS
namedtoasternamedtoaster/The-Last-Mile/skills/research/SKILL.md

research

Skill39 starsChanged 23 days agoMITView on GitHub

Answer a question by searching the project's own docs and code first, then official sources, and file the dated, sourced answer where it will be read again. Adds to an existing doc before creating a new one, marks anything unverified, and keeps a one line summary with the headline finding in the rules file's Docs index. Gives a short answer in chat first. Use when the user asks to research, look into, compare options for, or find out how something works before deciding or building.

189 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use daily-arxiv, exp-eval and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • LambenthanSkill

    daily-arxiv

    每日从 arXiv 拉取新论文,过滤相关性,auto-ingest 高优先级论文,检测 SOTA 更新

    Also in 5 projects that use this

  • LambenthanSkill

    exp-eval

    实验判决门:Review LLM 独立评判实验结果 → 4 种判决路径 → 自动更新 claims confidence、ideas status、graph edges

    Also in 5 projects that use this

  • truongtv22Skill

    Problem-Solving Techniques

    Apply systematic problem-solving techniques for complexity spirals (simplification cascades), innovation blocks (collision-zone thinking), recurring patterns (meta-pattern recognition), assumption constraints (inversion exercise), scale uncertainty (scale game), and dispatch when stuck. Techniques derived from Microsoft Amplifier project patterns adapted for immediate application.

    Also in 5 projects that use this

  • liufeng1976Skill

    setup-ts-deep-modules

    Wire dependency-cruiser into a TypeScript repo so each package is a deep module — implementation hidden in subfolders, reachable only through its entry-point files. User-invoked.

    Also in 43 projects that use this

  • skyllwtSkill

    exp-pilot-eval

    Pilot result evaluation — read pilot results, apply success criteria, update idea page (pilot_result, failure_reason if failed), generate PILOT_VERDICT_REPORT. Called by /ideate Phase 5 after /exp-pilot-run.

    Also in 4 projects that use this

  • skyllwtSkill

    exp-pilot-run

    Pilot experiment execution — read Pilot Spec YAML, write pilot code, run experiment(Confirm with the user before operation and require the applicant to conduct manual inspection), return results. Called by /ideate Phase 5. Does NOT modify wiki pages or judge pass/fail.

    Also in 4 projects that use this

  • +5 more from Lambenthan
  • +1 more from skyllwt

The file

Show its rules and text

We split this skill into the rules it gives an agent

22 rules. 0 are common, 20 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 169 edited versions we read keep these.

  • 20

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not write app code during research.

    line 16
  2. Distinctive:

    Before searching the web:

    line 27
  3. Distinctive:

    Read the Docs section of the rules file (CLAUDE.md or AGENTS.md).

    line 29
  4. Very specific:

    Search the docs folder and the root docs for the topic and its synonyms: grep -ril "<term>" docs/ *.md.

    line 30
  5. Distinctive:

    Search the code for the feature or library involved.

    line 32
  6. Distinctive:

    If an existing doc already answers the question and its date is recent enough for the topic, answer from it, cite it, and stop.

    line 34
  7. Distinctive:

    If a claim comes only from a secondary source, or from memory, mark it

    line 52
  8. Distinctive:

    Prefer concrete numbers: limits, prices, sizes, timings, version numbers.

    line 54
  9. Distinctive:

    When sources disagree, say so and say which one you trust and why.

    line 55
  10. Distinctive:

    Prefer depth on the two or three real options over a long shallow list.

    line 56
  11. Distinctive:

    Standing decisions: if the user makes a decision from it, the rules file's

    line 70
  12. Distinctive:

    Gotchas: if it explains a surprise that cost time.

    line 71
  13. Distinctive:

    Never create an archive folder.

    line 76
  14. Distinctive:

    Keep it as short as the answer allows.

    line 91
  15. Distinctive:

    If the entry went into a doc that is already indexed (the backlog, the algorithms doc), update that doc's line only if its headline changed.

    line 107
  16. Very specific:

    If the rules file has no Docs section, add one at the end with the heading ## Docs and the sentence "One line per doc, with its headline …

    line 110
  17. Distinctive:

    Do not paste the whole write up into chat.

    line 123
  18. Distinctive:

    Do not create a new doc when an existing one covers the topic.

    line 128
  19. Distinctive:

    Do not leave an entry without a date or without sources.

    line 129
  20. Distinctive:

    Do not present a blog post, a forum answer or memory as confirmed.

    line 130
  21. Distinctive:

    Do not leave the Docs index line as a bare title.

    line 131
  22. Distinctive:

    Do not commit. The session end skill, or the user, does that.

    line 132

“Common” is measured: kept by at least 6 in 10 of the 169 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Research

Answer the question, then file the answer where someone will find it next time.

Most research docs are written once and never opened again. In one shipped app, 108 of 134 research docs were never touched after the day they were written. What got read was the one line summary in the rules file's Docs index, because the rules file is loaded every session. So the index line matters more than the doc. Write both, and write the index line as if it is the only thing anyone will read.

Do not write app code during research. If the answer says to change something, say so and let the user decide.

1. Pin down the question

Restate the question in one sentence, and the decision it should inform. If the question is too vague to answer (for example "look into payments"), ask one short question to narrow it. Otherwise start.

2. Look at what the project already knows

Before searching the web:

  1. Read the Docs section of the rules file (CLAUDE.md or AGENTS.md).
  2. Search the docs folder and the root docs for the topic and its synonyms: grep -ril "<term>" docs/ *.md. Try two or three words, not one.
  3. Search the code for the feature or library involved.

If an existing doc already answers the question and its date is recent enough for the topic, answer from it, cite it, and stop. Say how old it is. Store rules, prices and library versions go stale in months; product reasoning lasts longer.

3. Research

Sources, in this order of trust:

  1. Official docs for the thing in question (the store, the platform, the library's own docs and changelog, the vendor's pricing page).
  2. Primary sources: the library's source code or issue tracker, a release note, a test you run yourself.
  3. Secondary sources: blog posts, forum answers, videos, other people's skills and repos. Treat these as leads to check, not as answers.

Rules:

  • Cite every claim that matters with a link and the date you read it.
  • If a claim comes only from a secondary source, or from memory, mark it (unverified) and say what would confirm it.
  • Prefer concrete numbers: limits, prices, sizes, timings, version numbers.
  • When sources disagree, say so and say which one you trust and why.
  • Prefer depth on the two or three real options over a long shallow list.
  • For store rules or console steps, read the current official page now. Never state them from memory.

4. Decide where it goes

File into an existing doc whenever one fits. Create a new doc only when nothing does. Check in this order:

  1. An existing doc on the same topic found in step 2. Add a dated section, or update the old one and note what changed and when. Do not leave two answers that disagree.
  2. A doc the answer naturally belongs in: the backlog item it informs (add research notes to that item), the algorithms doc for logic, the rules file's Standing decisions if the user makes a decision from it, the rules file's Gotchas if it explains a surprise that cost time.
  3. A new doc, only if nothing above fits. Use templates/research-doc.md from this skill's folder. Put it in docs/research/ with a plain topic name, for example docs/research/push-notification-providers.md.

Never create an archive folder. Git keeps history.

5. Write the entry

Whether it is a new doc or a section in an old one, every entry has:

  • A date (YYYY-MM-DD) on the entry itself, not just the file.
  • The question, in one sentence.
  • The answer first: the finding and the recommendation, in two or three lines.
  • The detail: options, numbers, trade offs, rough effort (small, medium, large).
  • Sources, official first, each with a link and the date read.
  • Unverified: anything not confirmed, listed plainly with what would confirm it.
  • Open questions: what is still unknown and who could answer it.

Keep it as short as the answer allows. A long doc nobody reads helps nobody.

6. Update the Docs index

In the rules file's Docs section, add or update one line for the doc:

- `docs/research/push-notification-providers.md`: Provider A is cheapest under 10k
  users and has an Expo plugin; Provider B only wins on rich media (2026-10-11)

The line must carry the headline finding, not just the topic. "Research on push providers" is useless; the reader still has to open the file. Include the key number or the recommendation and the date. If the doc already has a line, rewrite it so it matches the newest finding.

If the entry went into a doc that is already indexed (the backlog, the algorithms doc), update that doc's line only if its headline changed.

If the rules file has no Docs section, add one at the end with the heading ## Docs and the sentence "One line per doc, with its headline number, so the summary is enough most of the time."

7. Answer in chat

Lead with the answer:

  1. The short answer and recommendation, in two to four lines.
  2. Anything unverified, in one line.
  3. Where it was filed: the doc path, and that the Docs index line was added or updated.

Do not paste the whole write up into chat. If the user wants the detail, it is in the doc.

Do not

  • Do not create a new doc when an existing one covers the topic.
  • Do not leave an entry without a date or without sources.
  • Do not present a blog post, a forum answer or memory as confirmed.
  • Do not leave the Docs index line as a bare title.
  • Do not commit. The session end skill, or the user, does that.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.