deep-research

Read this skill when the user asks for a research report, deep research, industry report, analysis report, whitepaper, or any substantive document deliverable; or when the topic requires multi-source retrieval and cross-validation (competitive analysis, market analysis, industry research, deep dive, landscape survey); or when the user is unsatisfied with the depth of an existing answer. Keyword signals: research, deep research, deep dive, analysis, investigate, report, survey, industry analysis, competitive analysis, market analysis. Trigger when: (1) the user wants to produce a formal report or research document as the deliverable; (2) the topic requires current, verifiable data from multiple sources — news, market analysis, competitive landscape, industry research; (3) the user is unsatisfied with the depth of an existing answer. Skip when a single search or model recall suffices and no document output is needed.

Quality

85%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Quality

Content

75%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A well-structured, actionable instruction skill with a clear gated workflow and decent progressive-disclosure scaffolding. The main gaps are mild redundancy across the Q&A/pseudoflow sections and the absence of explicit validation/recovery steps in the main execution phase.

Suggestions

Add an explicit validation/recovery loop to the main execution phase (e.g., 'if sources contradict or evidence is thin, re-run retrieval on the weak area before drafting'), mirroring the gap-diagnosis loop already present in deepen mode.

Consolidate the kickoff Q&A structure so it is defined once and referenced from both the new-research and deepen sections, and drop or shrink the 'Minimal Execution Pseudoflow' if it merely restates the numbered steps.

Add deepen-guide.md to the central 'References (Read On Demand)' list alongside research-workflow.md and report-template.html so all referenced files are discoverable in one place.

Dimension	Reasoning	Score
Conciseness	Mostly lean and assumes competence, but the kickoff Q&A structure is restated across Step 3 and the deepen section, and the 'Minimal Execution Pseudoflow' re-summarizes a workflow already described in the numbered steps.	4 / 5
Actionability	Gives concrete directives — 'Run one web_search call with 2-3 queries in parallel from different angles', specific files to read, a 3-7 question breakdown — but several steps remain high-level ('execute full workflow -> deliver').	4 / 5
Workflow Clarity	Clear Step 1→4 sequence with an explicit user-confirmation checkpoint before heavy execution and a gap-diagnosis feedback loop in deepen mode; the main execution phase, however, lacks explicit validation/recovery steps.	4 / 5
Progressive Disclosure	Good condition-mapped 'References (Read On Demand)' section that is one level deep and clearly signaled, but deepen-guide.md is referenced inline in the deepen section rather than in the central references list, and no bundle files are actually present.	4 / 5
	Total	16 / 20 Passed

Description

95%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A strong, explicit description that crisply defines the trigger surface, deliverable, and boundary against simpler alternatives. The only soft spot is specificity, where the named actions are methodological rather than concrete-tactile.

Dimension	Reasoning	Score
Specificity	Lists several concrete actions — 'multi-source retrieval and cross-validation', 'produce a formal report or research document as the deliverable', 'current, verifiable data from multiple sources' — plus comprehensive deliverable types, though the actions are methodological rather than tactile like the PDF anchor examples.	4 / 5
Completeness	Explicitly answers both what (produce verified, evidence-backed report deliverables via multi-source retrieval and cross-validation) and when (three numbered 'Trigger when' conditions plus a 'Skip when' boundary), with concrete trigger phrases throughout.	5 / 5
Trigger Term Quality	Comprehensive natural-term coverage with explicit synonyms in the 'Keyword signals' list — 'research, deep research, deep dive, analysis, investigate, report, survey, industry analysis, competitive analysis, market analysis' — matching how users actually ask.	5 / 5
Distinctiveness Conflict Risk	Clear niche — formal report/research document as deliverable — with an explicit 'Skip when a single search or model recall suffices' boundary that sharply reduces overlap with simpler search skills.	5 / 5
	Total	19 / 20 Passed

Validation

93%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 15 / 16 Passed

Validation for skill structure

Criteria	Description	Result
relative_links	Relative link issues: 2 missing	Warning

	Total	15 / 16 Passed

Repository: dtyq/magic
Path: backend/super-magic/agents/skills/deep-research/SKILL.md
Commit: f9973c5

Reviewed: 2 days ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.