Przejdź do treści
Biblioteka skilli
Nauka i badania · Research

perplexity-research

/perplexity-research

Wykonuje badania techniczne i dziedzinowe przez Perplexity MCP, gdy zadanie utknęło po co najmniej dwóch nieudanych próbach lub gdy trzeba je uruchomić ręcznie, ze statystykami wywołań.

Licencja
MIT
Ostatnia zmiana repozytorium
06.10.2026

Ocena i bezpieczeństwo

100/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść35 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
3
Skrypty
0
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 06.10.2026, bez odwołań do wycofanych modeli i API.

Licencja

MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.

Co robi i kiedy

Wykonuje badania techniczne i dziedzinowe przez Perplexity MCP, gdy zadanie utknęło po co najmniej dwóch nieudanych próbach lub gdy trzeba je uruchomić ręcznie, ze statystykami wywołań.

Oryginalny opis (EN)

Invoke when stuck (2+ failed attempts, uncertainty, or guessing) OR manually to research technical/domain questions via Perplexity MCP. ALWAYS announce invocation and track stats.

Jak działa

  1. When to Invoke (Automatic Triggers)
  2. Invocation Protocol
  3. Cost Efficiency
  4. Key Principles
  5. "I'm Stuck" Escalation Path
  6. References
  7. Companion Skills
Sekcje z nagłówków SKILL.md, w oryginale · 126 linii, 761 słów
Podgląd SKILL.md
---
name: perplexity-research
disable-model-invocation: true
source: superpowers-plus
augment_menu: true
triggers: ["/sp-research", "research this", "use perplexity", "I'm stuck", "need to research", "look this up", "stuck:research", "stuck:knowledge"]
anti_triggers: ["incorporate research", "merge research into doc", "add research findings"]
description: Invoke when stuck (2+ failed attempts, uncertainty, or guessing) OR manually to research technical/domain questions via Perplexity MCP. ALWAYS announce invocation and track stats.
summary: "Use when: stuck after 2+ failed attempts or need technical research via Perplexity."
coordination:
  group: stuck-escalation
  order: 2
  requires: []
  enables: []
  escalates_to: []
  internal: false
composition:
  consumes: [challenge]
  produces: [research-results, web-findings]
  capabilities: [searches-web, synthesizes-sources]
  priority: 10
---

# Perplexity Research

> **Purpose**: Get unstuck by dispatching research queries to Perplexity AI
> **Trigger**: Automatic (2+ failures, uncertainty) OR manual invocation
> **Cost**: Perplexity API calls are paid — use judiciously

## When to Invoke (Automatic Triggers)

You MUST invoke this skill when ANY of these conditions are met:

| Trigger | Description |
|---------|-------------|
| **2+ Failed Attempts** | Same operation failed twice with different approaches |
| **Uncertainty/Guessing** | You're unsure and would otherwise guess at an answer |
| **Cutting Corners** | About to do something that violates AGENTS.md guidance |
| **Hallucination Risk** | Making claims about APIs, libraries, or facts you're not certain about |
| **Outdated Knowledge** | Question involves tools/frameworks potentially released after training |
| **Unknown Errors** | Encountering error messages you can't interpret |

**Personal preference matters?** → Ask the user instead
**Broader/extrinsic research needed?** → Invoke this skill

## Invocation Protocol

User can force invocation with: "Use Perplexity to research X", "Get unstuck on X", "Research X via Perplexity".

### Step 0: Try Free Tools First (MANDATORY)

<EXTREMELY_IMPORTANT>
Try `web-search` + `web-fetch` first. Only escalate if ≥50% worse than expected.
State: "web-search returned [X]. Insufficient because [reason]. Escalating."
Never use Perplexity for: simple lookups, URL checks, basic fact-checking, reading docs.
</EXTREMELY_IMPORTANT>

### Step 1: Announce + Prompt

Announce: research topic, trigger, free tools tried, why escalating.
Prompt: context + specific question + constraints + what you've tried + web-search results.

### Step 2: Dispatch

| Quick fact | `perplexity_search_perplexity` | Deep research | `perplexity_research_perplexity` |
|------------|------|---------------|------|
| How-to | `perplexity_ask_perplexity` | Complex reasoning | `perplexity_reason_perplexity` |

### Step 3: Apply + Evaluate

Report findings → apply (run command/implement fix/test) → evaluate BEFORE recording stats.

### Step 6: Evaluate Helpfulness (CRITICAL)

After attempting to apply the Perplexity response, explicitly evaluate:

```markdown
📊 **Perplexity Evaluation**:
- Applied: [what you tried]
- Outcome: [SUCCESS | PARTIAL | FAILURE]
- Reason: [why it helped or didn't help]
```

**Evaluation criteria**:

| Outcome | Criteria | Record As |
|---------|----------|-----------|
| **SUCCESS** | Problem solved, unblocked, or gained actionable insight | `successful: true` |
| **PARTIAL** | Some useful info but needed additional work | `successful: true` |
| **FAILURE** | Information was wrong, irrelevant, or didn't help | `successful: false` |

### Step 7: Record Outcome (After Evaluation)

**ONLY after Step 6**, log the research outcome for future reference:

- **Trigger**: What caused the research (failed_attempts, uncertainty, etc.)
- **Tool used**: ask, search, or reason
- **Query summary**: Brief description
- **Outcome**: SUCCESS, PARTIAL, or FAILURE
- **Reason**: Why it helped or didn't

> Record outcomes in conversation context or TODO notes — this helps calibrate when Perplexity is worth the cost.

**The evaluation loop**:

1. Receive Perplexity response → Report (Step 4)
2. Apply the information → Act (Step 5)
3. Evaluate outcome → Judge (Step 6)
4. Record outcome → Track (Step 7)

| Failure | Fix |
|---------|-----|
| Using Perplexity when web-search would suffice — cost waste | Always complete Step 0 (free tools first) and state what they returned |
| Treating Perplexity response as authoritative without cross-verification | Perplexity can hallucinate too — verify key claims against primary sources |
| Not recording outcome, preventing future calibration | Always complete Step 7 (record outcome) even when the answer was unhelpful |
| Prompt too vague — getting generic response | Include context, constraints, what you tried, and what web-search found |

## Cost Efficiency

> ⚠️ **Perplexity API calls cost real money.** Always try free tools first (Step 0).
> See `references/cost-reference.md` for high/low-value use cases, efficiency tactics, and the full cost-conscious decision framework.

## Key Principles

1. **FREE FIRST** - Always try web-search and web-fetch before Perplexity
2. **ALWAYS announce** - User must know when Perplexity is being consulted
3. **Justify escalation** - State what web-search found and why it's insufficient
4. **Track everything** - Stats enable tuning and improvement
5. **Rich prompts** - Better prompts = better results
6. **Low threshold** - 2 failures is enough; don't struggle unnecessarily
7. **Cost awareness** - Perplexity costs real money; use only when free tools fail

## "I'm Stuck" Escalation Path

Default order: `think-twice` (free, reasoning) → `perplexity-research` (paid, knowledge).
See `references/escalation.md` for the full decision tree.

## References

- [`references/cost-reference.md`](references/cost-reference.md) — High/low-value use cases, efficiency tactics, cost-conscious decision framework
- [`references/escalation.md`](references/escalation.md) — "I'm stuck" decision tree for think-twice vs perplexity-research

## Companion Skills

- **expert-interviewer**: For structured domain expert interviews
- **incorporating-research**: Merging Perplexity findings into docs
- **think-twice**: When stuck — Perplexity is an escalation target
Pliki w paczce (3)
  • skill.md 6 KB
  • references/cost-reference.md 3 KB
  • references/escalation.md 1 KB

Wersje

RepozytoriumWynikLicencja
Ta strona · bordenet/superpowers-plus100MIT
garrytan/gbrain100MIT
dreamlab-ai/agentbox100AGPL-3.0
baofeng-tech/agent-skills100MIT

Migawka bazy z 13.11.2026 · wersja paczki 492cd1044eae (skrót SHA-256 całej paczki)

Pobrania w ostatnich 30 dniach: 0 · ranking

Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.

Podobne skille

  • 100perplexity
    Wyszukuje informacje w sieci i prowadzi research przez Perplexity AI przy ogólnych pytaniach o aktualności, z wyłączeniem dokumentacji bibliotek i pytań o workspace.
  • 100deep-research
    Prowadzi głęboki research przez DeepAPI (w tym Perplexity Deep Research) i zapisuje raport ze źródłami.
  • 100perplexity-common-errors
    Diagnozuje i naprawia typowe błędy i wyjątki Perplexity, w tym nieudane żądania i problemy integracji.
  • 100perplexity-search
    Wywołuje Perplexity Sonar, Sonar Pro, Sonar Reasoning Pro lub Sonar Deep Research przez AIsa dla odpowiedzi z cytowaniami, rozumowania analitycznego i długich raportów.
  • 100perplexity-ci-integration
    Konfiguruje integrację Perplexity z CI/CD w GitHub Actions oraz automatyczne testy w procesie budowania.
  • 90known-problem-hint-research
    Szuka użytecznego tropu badawczego, sprawdzając dokumentację, zgłoszenia i artykuły, gdy zadanie techniczne utknęło.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.