Description
Add a GitHub Copilot custom agent dedicated to long-form editorial/article work.
The agent should support a unified workflow rather than separate generation and polishing agents:
- generate a new article from a topic, brief, outline, or partial notes
- polish an existing draft written manually or produced elsewhere
- normalize structure and metadata so the result is ready for Renderflow templates/profiles
The goal is to make a rough idea or draft converge toward a consistent, publication-quality article without requiring the user to manually restate formatting and editorial expectations each time.
Location
Create the agent using GitHub's repository-level custom-agent convention:
.github/agents/medium-article.agent.md
Inspect the current repository's existing custom-agent conventions before implementing so this agent follows the same metadata and structure.
Goals
- provide one reusable Copilot role for both generation and polishing
- preserve the author's meaning and voice while improving clarity and structure
- produce deterministic, template-ready Markdown and metadata
- align generated content with Renderflow's long-form editorial template/profile
- avoid inventing facts, citations, quotations, credentials, or claims merely to make an article feel complete
Content Contract
The agent should be able to work from any of the following states:
- Topic or short brief only
- Loose notes or brain dump
- Partial article
- Complete rough draft needing polish
- Existing Markdown that needs normalization to the article contract
The resulting content should use an editorial structure appropriate to the source rather than forcing every article into an identical outline. It should support, where useful:
- title
- optional subtitle/dek
- concise opening/lede
- readable section hierarchy
- short, scannable paragraphs
- lists where they improve comprehension
- quotations only when source material actually supplies them
- figures/images with descriptive captions when supplied
- code blocks and technical examples when relevant
- conclusion, takeaway, or closing reflection when appropriate
- structured front matter required by the paired Renderflow template/profile
Generation Behavior
When the user supplies an idea, brief, or outline rather than a complete draft:
- build a coherent long-form article from the supplied material
- preserve user-provided claims and framing
- make uncertainty explicit rather than filling factual gaps with invented material
- use placeholders only when the missing information truly requires user input
- produce valid Markdown compatible with Renderflow
Polishing Behavior
When an existing draft is supplied:
- treat the draft as canonical source material unless the user requests substantive rewriting
- preserve the author's intent, tone, and distinctive phrasing where practical
- improve organization, flow, grammar, heading structure, and readability
- remove accidental repetition and filler without flattening the author's voice
- preserve technical details, caveats, citations, links, code, and source attribution
- normalize metadata and formatting to the article output contract
- do not silently strengthen claims or introduce new factual assertions
Template/Profile Alignment
The agent should understand that:
Copilot agent = content generation and editorial normalization
Renderflow template/profile = presentation and deterministic rendering
Do not encode visual styling details into article prose when those belong in the template/profile.
The agent output should be usable by the paired Medium-style/editorial Renderflow profile without manual structural repair.
Naming / Trademark Constraint
"Medium-style" refers to a clean, readable long-form web-article experience and workflow inspiration. Do not copy Medium's proprietary visual design, branding, assets, or protected product presentation.
Acceptance Criteria
Relationship to Existing Work
This complements the existing research-paper Copilot agent rather than replacing it. The research agent owns academic/research structure; this agent owns accessible long-form editorial/article structure.
Description
Add a GitHub Copilot custom agent dedicated to long-form editorial/article work.
The agent should support a unified workflow rather than separate generation and polishing agents:
The goal is to make a rough idea or draft converge toward a consistent, publication-quality article without requiring the user to manually restate formatting and editorial expectations each time.
Location
Create the agent using GitHub's repository-level custom-agent convention:
Inspect the current repository's existing custom-agent conventions before implementing so this agent follows the same metadata and structure.
Goals
Content Contract
The agent should be able to work from any of the following states:
The resulting content should use an editorial structure appropriate to the source rather than forcing every article into an identical outline. It should support, where useful:
Generation Behavior
When the user supplies an idea, brief, or outline rather than a complete draft:
Polishing Behavior
When an existing draft is supplied:
Template/Profile Alignment
The agent should understand that:
Do not encode visual styling details into article prose when those belong in the template/profile.
The agent output should be usable by the paired Medium-style/editorial Renderflow profile without manual structural repair.
Naming / Trademark Constraint
"Medium-style" refers to a clean, readable long-form web-article experience and workflow inspiration. Do not copy Medium's proprietary visual design, branding, assets, or protected product presentation.
Acceptance Criteria
.github/agents/medium-article.agent.mdexists and follows current GitHub custom-agent conventionsRelationship to Existing Work
This complements the existing research-paper Copilot agent rather than replacing it. The research agent owns academic/research structure; this agent owns accessible long-form editorial/article structure.