Menú
Systematic 4-phase research methodology: Scope → Gather → Synthesize → Deliver. Single source of truth — no /commands wrapper.
Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.
Basado en la clasificación ocupacional SOC
Review existing Architecture Decision Records. Flag ADRs whose aging signals may have triggered given current state. Categorizes each as valid / aged / superseded and suggests follow-up actions.
Author an Architecture Decision Record (ADR). Captures decisions with implications beyond the current change — lasts months, affects multiple files, hard to reverse. Records context, alternatives considered, consequences, and aging signals so the decision can be revisited when conditions change.
4-phase loop-detection and recovery protocol. Fires when an agent detects it is stuck or repeating the same tool call with no progress — before tool-loop-detection circuit-breaks at 30 calls.
Socratic exploration of requirements before implementation. Refines spec through targeted questions. Use when user asks to 'explore options', 'refine requirements', or mentions 'unclear requirements'.
Generate structured changelog from git history. Groups commits by type, filters noise, produces Keep a Changelog format. Single source of truth — no /commands wrapper.
Generate pipe-delimited commits: Tipo|IdTarea|YYYYMMDD|Descripción. Single source of truth — no /commands wrapper.
| name | deep-research |
| description | Systematic 4-phase research methodology: Scope → Gather → Synthesize → Deliver. Single source of truth — no /commands wrapper. |
| version | 2.0.0 |
| when | {"keywords":["research","investigate","explore","compare","analyze options","deep-research"]} |
| allowed-tools | ["WebSearch","WebFetch","Grep","Read","Bash(rg *)","Bash(fd *)"] |
Systematic research for technical and product questions. Use when external source validation is required.
Before searching, write:
RESEARCH QUESTION: [exact question]
SCOPE: [in / out of scope]
SUCCESS CRITERIA: [what does a good answer look like?]
KNOWN CONTEXT: [what we already know]
UNKNOWNS: [what specifically needs verification]
Do NOT start searching until this is written.
WebSearch("query with current year 2026")
WebFetch(url, "extract [specific information needed]")
Grep("pattern", path)
| Question | Answer | Source | Confidence |
|---|---|---|---|
| [q1] | [answer] | [url/file] | high/medium/low |
List contradictions discovered. Explain which source to trust and why.
List what could NOT be verified.
## Research: [Topic]
### Bottom Line
[Conclusion first — 2-3 sentences max]
### Evidence
[Findings table from Phase 3]
### Recommendation
[Concrete action based on evidence]
### Caveats
[Confidence gaps, conflicting sources, what might change]
### Sources
- [Source 1](url) — [why trusted]
- [Source 2](url) — [why trusted]
| Anti-pattern | Correct behavior |
|---|---|
| Searching without a question | Define question first (Phase 1) |
| Trusting single source | Cross-reference 2+ authoritative |
| Fabricating citations | Only cite sources actually fetched |
| Confidence without evidence | Use confidence levels |
| "Typically" / "usually" | Find docs and cite |
| Outdated info | Note pub date, flag if > 2 years |
Skill('deep-research') topic="<research question>"