staruhub/ClaudeSkills / skills/Geek-skills-deep-research
deep-research
VerifiedUse this skill when the user wants an evidence-based research memo, literature review, market/policy/technical landscape, or a multi-source decision brief with citations, trade-offs, and a clear conclusion. Best for tasks that need synthesis across multiple external sources, iterative follow-up research, or a reusable written artifact. Do not use for quick factual lookups, single-source summaries, simple Q&A, summarizing one document the user already provided, plan-only requests where the user explicitly defers the actual research, or when the user wants a short answer with no need for cited evidence (a short but evidence-backed conclusion is still a brief memo, which this skill covers). Chinese trigger examples: "帮我调研", "深度研究", "综述报告", "技术选型分析", "竞品研究", "政策分析".
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 11.0 KB |
| assets/report_template.md | markdown | 1006 B |
| evals/routing-evals.json | config | 3.8 KB |
| evals/runbook.md | markdown | 967 B |
| references/evaluator-prompt.md | markdown | 4.0 KB |
| references/handoff-format.md | markdown | 3.9 KB |
| references/landscape-scan.md | markdown | 1.2 KB |
| references/methodology.md | markdown | 5.7 KB |
| references/observability.md | markdown | 3.0 KB |
| references/quality-gates.md | markdown | 5.5 KB |
| references/report-assembly.md | markdown | 3.4 KB |
| references/research-notes-format.md | markdown | 2.2 KB |
| references/subagent-prompt.md | markdown | 1.7 KB |
| references/tension-discovery.md | markdown | 1.2 KB |
| scripts/emit_run_summary.py | script | 5.0 KB |
| scripts/source_evaluator.py | script | 11.9 KB |
| scripts/verify_citations.py | script | 10.7 KB |
references/observability.md
View on GitHub# Observability and Optimization V8
A good research skill should leave a structured trace behind.
Narrative logs are nice; structured summaries are what let you debug regressions.
## Emit `run-summary.json`
Use `scripts/emit_run_summary.py` after drafting/evaluation.
### Minimum fields
- `skill_version`
- `timestamp`
- `output_type`
- `stakes`
- `orchestration_mode`
- `used_subagents`
- `used_tension_discovery`
- `used_landscape_scan`
- `used_reverse_search`
- `search_count` / `fetch_count` (if known)
- `source_count`
- `citation_count`
- `word_count`
- `evaluation_run`
- `evaluation_verdict`
- `issues`
### Recommended fields
- `estimated_effort`
- `latency_seconds`
- `token_in` / `token_out` (if your harness can provide them)
- `manual_spotcheck_run`
- `accepted_without_major_edits`
- `failure_class`
---
## What to observe in traces
If your platform supports tracing, capture these events:
- skill activated / not activated
- output type chosen
- single-agent vs subagents
- each search / fetch
- each script call
- compaction / context reset
- evaluation run and result
- publish / stop / escalate
This makes it easier to answer:
- Why did the skill trigger?
- Why did it get expensive?
- Where did the run lose grounding?
- Which optional module actually helped?
---
## Core KPIs
### Routing KPIs
- trigger precision
- trigger recall
- over-trigger rate
- under-trigger rate
### Process KPIs
- notes completeness
- registry completeness
- citation integrity pass rate
- evaluation pass rate
### Outcome KPIs
- user accept / minimal-edit rate
- groundedness score
- coverage score
- usefulness / actionability score
### Efficiency KPIs
- median time per run
- median tool calls per run
- median tokens per run
- cost per accepted artifact
---
## Failure taxonomy
When a run fails, classify it before editing the skill.
### 1. Routing failure
Examples:
- skill should not have triggered
- full report chosen where brief memo was enough
### 2. Process failure
Examples:
- missing notes
- no counter-evidence search
- subagents duplicated work
### 3. Grounding failure
Examples:
- weak support for key claims
- citation mismatch
- wrong source type for the claim
### 4. Output failure
Examples:
- did not answer the actual question
- too generic
- forced contrarianism
### 5. Efficiency failure
Examples:
- too many searches
- unnecessary subagents
- evaluation run not worth the cost
### 6. Governance failure
Examples:
- violated data handling boundary
- missing required human review
Fix the **smallest layer** that caused the failure.
Do not rewrite the whole skill when the real bug is just the description or one quality gate.
---
## Optimization loop
1. Collect failed or expensive runs
2. Classify each failure
3. Add or update a routing or output eval case
4. Change one thing only
5. Re-run the affected eval slice
6. Compare against the previous version
7. Keep the change only if the metrics improve or at least do not regress badly
This is how a skill becomes more reliable instead of just longer.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |