leszekgruchala/buddy / skills/research
research
VerifiedResearch code, behavior, docs, flows, comparisons, and technical facts without recommending or choosing changes; change decisions belong in spec. Keep the current request in this skill; route new requests by their own scope.
Research
Keep the current request in this skill. Do not activate another Buddy skill without user or develop direction. Workers return to their caller. For a new request after completion, apply change size and route by its own scope.
Resolve .ai/ paths from the user's workspace root (its git root), never from the skill or plugin directory.
Gather facts only. Do not recommend, decide, or implement.
Model selection
Apply model selection; recommend balanced.
Rules
- Read code and current primary documentation; make no working-tree changes outside the required research artifact and its
trash/. - Separate verified facts, inferences, and unknowns.
- Stop at the depth bar, not at the first plausible answer; omit unneeded implementation detail.
- Ask only when the answer cannot be discovered safely.
- Choices, preferences, and action-oriented follow-ups such as "let's do X" update the same research artifact; they do not authorize a spec or implementation. An explicit user transition or the calling
developorchestrator selects the next skill; in adeveloprun, its judge rules on choices. Remain in research when intent is ambiguous.
Depth
Scale to the work: a factual lookup and trivial or small work need steps 1, 2, and 5; substantial work or an unsettled approach needs all five.
- Test the premise. Check what the request, caller, or earlier artifact assumes: that the problem exists, its stated cause, the named component, and that the question asked is the one that decides the work. When a premise is false or the real problem lies elsewhere, say so in
OUTCOMEand follow the real problem; do not answer only the question as framed. - Reach primary evidence. Trace each claim to the code that runs, its tests, observed behavior, or current official documentation. Names, comments, READMEs, summaries, and earlier artifacts are leads, not proof. Follow a symptom to its cause, and run a safe read-only check when reading leaves doubt.
- Try to disprove the answer. Look for what would make it wrong, including other callers, configuration, and the history behind the code.
- Map the solution space as facts. Find what already solves or nearly solves the problem in the repository, its dependencies, and its platform; how current official sources and established practice solve it; and which constraints rule an approach out. Do not rank or choose.
- Do not give up at the first dead end. A failed search is not an unknown. Before recording an unknown that could change an outcome, try at least two other angles: other names, callers, tests, git history, configuration, official documentation, or a run.
Stop when every question has a direct outcome on primary evidence, the premise is settled, any attempt to disprove found nothing new, and further lookups no longer change an outcome. Drop a line of inquiry when its answer cannot change an outcome or the next stage.
Research artifact
Resolve the path at the start. Reuse the relevant passed worklog and research artifact for follow-ups; otherwise use .ai/worklog/<yyyyMMdd>_<work-name>/research_<work-name>.md and .ai/worklog/<yyyyMMdd>_<work-name>/trash/ for temporary files. A follow-up that asks a different question about the same work adds a question to the same artifact. Do not create, hand off, or leave a scaffold-only research file. Write on material findings and update material conclusion changes. Before any user-facing handoff, record the questions, direct outcomes, evidence-backed findings, and unknowns concisely. Put the verified authoring runtime model slug in top YAML front matter under model_slug, following model selection; never use inherit, a tier, unresolved alias, or unverified requested model.
Keep the artifact current
The artifact records what is true and relevant now, not how the research got there. Decide what to keep by evidence and relevance to the work, never by which question is newest.
On every update, reconcile the whole artifact before writing:
- Check new evidence against every existing outcome, finding, unknown, and decision, not only those of the current question.
- Correct or delete each item that evidence contradicts or invalidates. Do not keep it as history, strike it through, or mark it as earlier or previous. Record a ruled-out explanation or option only in
REJECTED, and only when it prevents repeated investigation. - Keep every question whose answer still affects the work, with its outcome, findings, and evidence, also when the user asks a different question. A new, narrower, or related question does not supersede an earlier one. A question is superseded only when it is answered and no longer affects the work, evidence invalidates its premise, or the user drops it.
- Remove resolved unknowns; keep their answers as findings when they still matter. Merge duplicates. Do not append dated follow-ups, discussion transcripts, or earlier versions of an answer.
- If nothing material changed, do not write.
In the handoff, state in one line each material outcome, finding, question, or constraint that the update removed or reversed, and why. Do not report routine compression.
Keep the artifact short
- There is no word limit: length follows current, relevant content. Never pad, and never drop, defer, or refuse to record still-relevant content to save words.
- Per question, use one sentence for the question, one to three for its outcome, and only the short finding bullets that support it. Label items
Qnand keep numbers stable; omit labels when there is one question. Use a compact table only when it makes a comparison easier to read. Keep evidence links beside claims; label inferences and retain material limitations. Link to source detail instead of copying it. - Keep only unresolved questions that affect an outcome or the next stage in
UNKNOWNS, each with what was checked and what would settle it; writeNonewhen there are none. Exclude investigation chronology, tool logs, session housekeeping, repeated caveats, implementation plans, and unrelated background. - Add
PAST DECISIONSonly for earlier user or orchestrator decisions and constraints that still apply or that prevent repeated exploration: at most three one-sentence bullets stating each and its brief reason. Record supplied decisions, not new research recommendations. Remove obsolete entries; revisit only when the user reopens them or new evidence invalidates their premise. - Add
REJECTEDlast, only for ideas, options, or explanations that the user, the orchestrator, or evidence ruled out and that would otherwise be revisited: one line each, naming it with the reason and evidence link. Keep no other trace of a rejected item in the artifact. Remove an entry once it no longer prevents repeated exploration; move it back only when the user reopens it or new evidence invalidates the reason. - Before every write and handoff, read the whole artifact, reconcile it as above, and tighten wording. Never drop a still-relevant question, outcome, finding, material uncertainty, or applicable constraint.
---
model_slug: <exact runtime model slug>
---
# <research-name>
## QUESTIONS
- Q1: <question that still affects the work, in one sentence>
## OUTCOME
- Q1: <direct conclusion>
## FINDINGS
- Q1: <essential fact or labeled inference with its evidence link>
## UNKNOWNS
- Q1: <material unresolved question>; checked: <angles tried>; settled by: <missing evidence>, or None
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-changelog-updatemorluto/reaPrepare or rewrite REA release changelogs and GitHub release notes from pinned Git history, with verified contributor thanks and Release Please synchronization. | Stars: 470k | Updated | Claude CodeCodex | Verified | Install |
| academy-guideanthropics/skillsStop and check this skill before finishing any reply to a question about how to use Claude or a Claude product — it recommends matching courses, tutorials, and use cases from Claude Academy (academy.claude.com), Anthropic's learning hub. Trigger on: "how do I", "how can I", "getting started with", "what can Claude do", "teach me", "learn to use"; questions about artifacts, projects, skills, plugins, connectors, MCP; requests about rolling Claude out to a team, class, or organization; and any ask for training materials, onboarding content, or learning resources. Use it when the user is learning how to use a feature or product — not when they are mid-task and just want the task done. This skill composes with other skills: after consulting product documentation to answer how a Claude feature works, also check here for a matching course or tutorial — a docs-grounded answer and an Academy recommendation belong together. Only recommend on a strong match; never invent Academy content. | Stars: 197k | Updated | Claude CodeCodex | Verified | Install |
| claude-apianthropics/skillsReference for the Claude API / Anthropic SDK — model ids, pricing, params, streaming, tool use, MCP, agents, caching, token counting, model migration. TRIGGER — read BEFORE opening the target file; don't skip because it "looks like a one-liner" — whenever: the prompt names Claude/Anthropic in any form (Claude, Anthropic, Fable, Opus, Sonnet, Haiku, `anthropic`, `@anthropic-ai`, `claude-*`, `us.anthropic.*`, `[1m]`); the user asks about an LLM (pricing/model choice/limits/caching) — never answer from memory; OR the task is LLM-shaped with provider unstated (agent/MCP/tool-definition/multi-agent/RAG/LLM-judge/computer-use; generate/summarize/extract/classify/rewrite/converse over NL; debugging refusals/cutoffs/streaming/tool-calls/tokens). SKIP only when another provider is being worked on (overrides all triggers): OpenAI/GPT/Gemini/Llama/Mistral/Cohere/Ollama named in the query; OR `grep -rE 'openai|langchain_openai|google.generativeai|genai|mistralai|cohere|ollama'` over the project hits (run this grep FIRST if no provider named — don't Read the file). | Stars: 197k | Updated | Claude CodeCodex | Unverified | Install |
| doc-coauthoringanthropics/skillsGuide users through a structured workflow for co-authoring documentation. Use when user wants to write documentation, proposals, technical specs, decision docs, or similar structured content. This workflow helps users efficiently transfer context, refine content through iteration, and verify the doc works for readers. Trigger when user mentions writing docs, creating proposals, drafting specs, or similar documentation tasks. | Stars: 197k | Updated | Claude CodeCodex | Verified | Install |