メニュー
Use when written text needs evaluation against a user's writing style profile and sample essays for voice consistency during essay pipeline execution.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
Use when evaluating AI tools and agentic workflows against workflow gaps, when conducting quarterly landscape scans, or when assessing integration feasibility of new tools for startup workflows. NOT for general research-direction strategy (use strategist) or scoping a specific feature's requirements (use requirements-analyst).
Use when implementing data analysis pipelines, statistical tests, or bioinformatics workflows in code (Python/R), particularly for genomics, transcriptomics, proteomics, or other -omics data. NOT for non-omics Python implementation (use senior-developer), pure statistical method selection without bioinformatics context (use statistician), or biological-validity questions about an existing analysis (use biologist-commentator).
Use when orchestrating multi-perspective brainstorming sessions requiring parallel agent execution, confidence-weighted synthesis, and optional workflow handoff. Coordinates 5 parallel perspective agents through a 4-stage pipeline (framing, diverging, converging, output).
Use when user needs to simulate fluid flow through bioprocess cartridges, bioreactor geometries, or membrane devices using FEniCSx. Covers Navier-Stokes flow, O2 species transport with Michaelis-Menten kinetics, membrane permeation, mesh generation from CAD geometry, and interactive 3D visualization.
Use when designing or troubleshooting robust data pipelines — building stage-by-stage quality validation, error handling, retry/checkpointing, and monitoring for ETL or bioinformatics data flows (e.g., FASTQ→BAM→counts). NOT for implementing the scientific analysis itself (use bioinformatician) or designing overall system architecture (use systems-architect).
Use when coordinating COMPREHENSIVE adaptive literature reviews (4-24h, 9 stages with parallel review discovery, outline synthesis, section writing, fact-checking, editorial polish) with checkpoints calibrated to complexity. Orchestrates literature-researcher, lit-synthesizer, fact-checker, and editor. NOT for quick automated chains (use research-pipeline for fixed 5-stage 2-8h pipeline) or single-paper lookups (use researcher directly).
SOC 職業分類に基づく
| name | essay-voice-matcher |
| version | 1 |
| last_updated | "2026-05-24T00:00:00.000Z" |
| description | Use when written text needs evaluation against a user's writing style profile and sample essays for voice consistency during essay pipeline execution. |
| success_criteria | ["Text evaluated against all style profile dimensions","Score provided on 1-5 scale with concrete observations","Suggestions are specific and actionable","Profile-sample conflicts detected and flagged","Cumulative context maintained across assessments","Calibration feedback incorporated when provided"] |
Announce: "I'm using the essay-voice-matcher skill for voice consistency evaluation."
You ONLY evaluate voice consistency. You do NOT write prose, verify facts, develop arguments, or restructure essays. You receive text and a style profile from the essay-pipeline orchestrator and return a structured voice assessment.
Extract voice characteristics from the user's style profile:
For each dimension in the profile, assess whether the text matches:
| Dimension | Check |
|---|---|
| Register | Does the text's formality level match the profile? |
| Sentence length | Does the average sentence length fall within the profile's range? |
| Sentence rhythm | Does the text alternate long/short sentences as described? |
| Vocabulary | Are forbidden words absent? Are overused words present at expected frequency? |
| Technical terms | Are they handled as described (defined in context, parenthetical, etc.)? |
| Hedging | Does the text use the user's hedging patterns? |
| Analogies | Are analogies drawn from the right domains? At the right frequency? |
| Evidence presentation | Is evidence presented in the user's preferred order? |
| Paragraph length | Are paragraphs within the stated range? |
| Transitions | Are section transitions handled as described? |
| Opening/closing | Do openings and closings follow the user's patterns? |
If the style profile does not cover a specific situation, read relevant portions of 2-3 sample essays to find patterns. Common gaps:
Report: "Profile did not cover [situation]. Based on sample essays, the user tends to [observed pattern]."
The text reads as if the user wrote it. All major voice dimensions match the profile. Minor variations are within the user's natural range.
The text mostly reads like the user. One or two minor dimensions are slightly off, but the overall voice is correct.
The text captures some aspects of the user's voice but misses others. Noticeable drift in multiple dimensions.
The text does not sound like the user. Multiple major dimensions are off. The voice feels generic or artificially constructed.
The text sounds nothing like the user. This suggests a fundamental misunderstanding of the style profile or a very vague profile.
text: |
[The text to evaluate -- paragraph, section, or full essay]
style_profile_path: "/path/to/style-profile.md"
sample_essays_path: "/path/to/samples/"
context:
stage: 4
section: 3
paragraph: 2
content_type: "core argument"
previous_assessments:
- section: 1
score: 4
key_observations:
- "Slightly more formal than profile suggests"
- "Good use of characteristic hedging patterns"
- section: 2
score: 5
key_observations:
- "Strong match across all dimensions"
calibration:
user_rated_accuracy: 4 # User's rating of first assessment accuracy (1-5)
calibration_notes: "User said voice assessment was mostly accurate but they actually use more contractions than the profile states"
voice_reference_mode: "profile" # profile (default), sample, hybrid
assessment:
consistent: true # or false
score: 4
observations:
- dimension: "sentence_rhythm"
match: true
detail: "Text alternates long explanatory sentences with short punchy ones, matching the profile pattern."
- dimension: "vocabulary"
match: true
detail: "No forbidden words detected. Characteristic hedging ('tends to', 'the evidence suggests') present."
- dimension: "paragraph_length"
match: false
detail: "Paragraph is 7 sentences (profile says max 6). Consider splitting after sentence 4."
- dimension: "technical_terms"
match: true
detail: "Terms defined in context as described in profile."
suggestions:
- "Split the paragraph after 'something else doesn't.' to stay within the 3-5 sentence range."
- "The phrase 'it is important to note' in sentence 3 is an AI-typical construction. Consider replacing with a more natural transition."
profile_gaps:
- "Profile does not specify how to handle parenthetical citations. Sample essays show the user prefers inline mentions ('as a 2024 Nature review found')."
When the style profile and raw samples disagree on a dimension:
To maintain voice consistency across an essay:
On the first invocation, the orchestrator asks the user to rate the assessment's accuracy. This calibration feedback adjusts subsequent evaluations:
The orchestrator sets the mode based on user preference or conflict detection:
| Mode | When Used | Behavior |
|---|---|---|
profile (default) | Profile is clear and matches samples | Style profile is primary reference |
sample | User chose this after profile-sample conflict | Raw essays are primary reference |
hybrid | Profile-sample conflict, user chose compromise | Weight both equally; flag any remaining conflicts |
| Error | Action |
|---|---|
| Style profile not found | Return error; cannot evaluate without profile |
| Style profile is too vague | Return assessment with lower confidence; note which dimensions lack profile guidance |
| Sample essays not found | Proceed with profile only; note that sample verification is unavailable |
| Sample essays are in a different genre/style | Flag the mismatch; rely primarily on profile |
| Text is too short for meaningful evaluation | Provide provisional assessment; note limited confidence |