ask-deep-research
Deep Research: Runs autonomous multi-source investigation with contradiction analysis and a cited handoff. Use for complex technical questions. Common triggers: deep research, exhaustive research, comprehensive investigation, complex technical investigation, complex contested high-stakes question, complex compatibility question, complex compatibility issue, complex question, contested research, high-stakes research, full compatibility investigation, compare competing implementations, compare local and upstream implementations, investigate historical changes, determine protocol behaviour, investigate protocol behavior, protocol behavior exhaustively, multiple sources, research everything relevant, investigate open issues, open issues comprehensively, compare against upstream.
- License 许可
- MIT
- 最近更新
- 2026-11-09
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- ask-deep-research
- description
- Deep Research: Runs autonomous multi-source investigation with contradiction analysis and a cited handoff. Use for complex technical questions. Common triggers: deep research, exhaustive research, comprehensive investigation, complex technical investigation, complex contested high-stakes question, complex compatibility question, complex compatibility issue, complex question, contested research, high-stakes research, full compatibility investigation, compare competing implementations, compare local and upstream implementations, investigate historical changes, determine protocol behaviour, investigate protocol behavior, protocol behavior exhaustively, multiple sources, research everything relevant, investigate open issues, open issues comprehensively, compare against upstream.
- disable-model-invocation
- true
02 资源文件 1
- 参考文档research-report-template.md1.8 KB
03 SKILL.md 正文
ASK Deep Research
Turn hard technical uncertainty into an autonomous research program. Gather and challenge evidence; do not implement source changes.
Start
Write a research brief before broad search: question, objectives and decisions enabled, in/out scope, constraints, key unknowns, initial hypotheses, expected report. Do not repeatedly ask permission when the next source follows from this brief.
Research Program
- Build a dynamic plan. Split independent tracks: local implementation, real fixtures/logs/captures, upstream, history, official standards/docs, and community evidence where relevant.
- Inspect local code, tests, fixtures, configuration, Git history, issues, and PRs before interpreting external claims about local behavior.
- Discover sources broadly, then narrow with identifiers, versions, hardware, protocol fields, and terminology learned. Prefer direct runtime evidence, reproducible tests, source, standards, and official docs.
- Delegate independent tracks through
agent-workflows; coordinator owns scope, evidence reconciliation, plan updates, and synthesis. Report meaningful findings at milestones, not tool noise. - Extract evidence, not URL lists:
Source | section | observation | interpretation | confidence | contradiction. - Track each important statement as
FACT,OBSERVATION,INFERENCE,HYPOTHESIS,ASSUMPTION, orUNKNOWN. Maintain hypotheses with support, contradictions, status (OPEN,SUPPORTED,WEAKENED,DISPROVEN,CONFIRMED), and confidence. - Compare working/broken, old/new, local/upstream, hardware/software variants, fixtures, timing, fields, mappings, and state transitions. For protocols inspect identity, addresses, IDs, lengths, positions, types, bits, enumerations, read/write semantics, and observed state changes.
- After every pass, update plan: learned facts, changed hypotheses, open questions, new leads, and highest-value next source. Emit a checkpoint for long work using research-report-template.md.
- Search deliberately for disconfirming evidence before conclusions. Resolve conflicts by source authority, scope, history, and reproducible behavior; never silently choose a source.
- Stop only when answer is confirmed, actionable despite remaining uncertainty, evidence-blocked, or at diminishing returns. State why and name minimum next evidence.
Depth
- Standard: 3-5 tracks, multiple source types, iterative synthesis.
- Comprehensive: 5-10 tracks plus history, differential analysis, contradiction search, and hypothesis tests.
- Exhaustive: explicit request or justified stakes; expand corpus and track all evidence gaps without inventing certainty.
Output
Use research-report-template.md. Cite every externally derived factual claim; local citations use path:line, fixture section, commit, issue, or PR. Include a compact continuation state so a follow-up such as “dig deeper into point 3” resumes rather than restarts.
Return ASK_WORKFLOW_PASS phase=RESEARCH only with cited findings, confidence, remaining gaps, and implementation handoff. Use ASK_WORKFLOW_BLOCKED phase=RESEARCH when essential primary evidence is unavailable.
Use with
intaketo classify scope before or after evidence changes the decisionagent-workflowsfor independent research tracks and evidence contractdebuggingfor root-cause testing after research establishes likely causesdevelopfor the separate implementation pass after research handoff
Avoid
- One-search answers, confirmation bias, or search snippets as proof
- Treating names as protocol semantics without observed behavior
- Coding while evidence is still the deliverable
04 同仓库更多 Skills
ask-write-skill
Write Skill: Creates and revises skills and guardrails from reusable workflow gaps. Use when making a skill or when agent behavior exposes a routing gap.
ask-verification
Verification: Proves work is fixed or ready and cleans up changed repos. Use before claiming something works or handing off.
ask-text-writing
Text Writing: Writes emails, posts, articles, bios, READMEs, and docs that sound human and avoid AI writing patterns. Use when drafting or rewriting prose.
ask-summary
Summary: Explains what changed and what is proven, walks through code, or re-pitches unclear output. Use when work is done or the user asks what changed.
ask-session-review
Session Review: Reflects on skill usage and gaps and files GitHub issues from bugs, findings, or follow-ups. Use after completing work.
ask-research
Research: Does bounded fact-finding across the repository, docs, code, or external sources. Use when a question needs evidence before an answer or decision.