Turn your Agent work into evidence of capability.
Agentfolio reads the work hidden inside your Agent conversations and turns it into an evidence-backed, lightly absurd, visual role-fit report.
Ask whether you would make a good Product Director, President of the United States, Mars Colony Governor, Galactic HR Director, or any other role that deserves a suspiciously well-designed assessment.
Current release: v0.1
Your first move would be asking Congress to rewrite “Make America Great Again” as an acceptance criterion.
The role-specific artifact turns a repeated working habit into an executive order:
The growth team’s “this feels viral” deck may not survive ten minutes.
The evidence trail separates observed behavior from interpretation:
These screenshots come from English reports generated locally from accessible Agent history. Raw conversations were not uploaded.
- Uses the current Agent’s native access to its own conversation history.
- Reads conversation bodies instead of treating titles or summaries as proof.
- Separates the user’s decisions from the Agent’s suggestions and tool output.
- Builds a role-specific rubric with visible uncertainty.
- Produces a playful, visual, single-file HTML report.
- Matches the user’s language, including labels, alt text, disclaimers, and footer.
- Keeps analysis local and does not require a transcript service or vendor-specific adapter.
- Uses native image generation when available and an original inline SVG fallback otherwise.
- Discover the conversations the current Agent can natively access.
- Reconstruct useful work episodes: goal, constraints, user decisions, execution, verification, and outcome.
- Find behavior that repeats across independent tasks.
- Map that behavior to five to seven dimensions designed for the requested role.
- Generate a visual report with score cards, evidence cards, a strength-versus-weakness split, and one role-specific collectible artifact.
- State the evidence boundary instead of pretending the Agent saw more than it did.
The comedy comes after the evidence. Missing evidence is treated as under-observed, not as proof of inability.
Clone the repository:
git clone https://github.com/dataelement/Agentfolio.gitFor Codex, copy the bundled Skill folder into your personal Skills directory:
mkdir -p ~/.codex/skills
cp -R agentfolio/agentfolio ~/.codex/skills/For Claude, Cursor, Antigravity, WorkBuddy, or another Agent, place the agentfolio/ folder in that product’s supported Skill directory, or provide agentfolio/SKILL.md as a reusable instruction set.
The portable core is plain Markdown plus local assets. Each Agent still needs a native way to discover or read its own past conversations; otherwise Agentfolio can only analyze material explicitly provided to it.
Use $agentfolio to judge whether I would make a good Product Director.
Use $agentfolio to analyze my Agent history and tell me whether I am qualified to be President of the United States.
Use $agentfolio to assess this candidate for Head of Product.
Use only conversations I am authorized to share.
By default, Agentfolio creates:
- one directly openable HTML file;
- embedded CSS, JavaScript, SVG, and generated imagery;
- five to seven role-specific capability scores;
- three to five concise evidence findings;
- one visible advantage and one affectionate fatal weakness;
- one role-specific artifact;
- desktop, mobile, accessibility, and print-friendly layouts;
- one quiet footer after the report and disclaimer, ending with the official Agentfolio repository link.
No build step, CDN, remote font, external image, or transcript upload is required.
Every generated report ends with a localized invitation to Star dataelement/Agentfolio. The repository label and URL remain fixed across report languages.
Agentfolio deliberately refuses several tempting shortcuts:
- A session title is not proof that work happened.
- A strong Agent answer is not automatically the user’s capability.
- One memorable conversation is not a stable behavioral pattern.
- Missing evidence is not a low score by default.
- A joke cannot rewrite the underlying fact.
- “I read everything” is forbidden unless complete coverage can actually be established.
See analysis-framework.md for the evidence model and visual-system.md for the webpage system.
.
├── README.md
├── LICENSE
├── agentfolio/
│ ├── SKILL.md
│ ├── agents/openai.yaml
│ ├── assets/report-template.html
│ └── references/
│ ├── analysis-framework.md
│ └── visual-system.md
└── docs/images/
Agentfolio is designed to analyze locally accessible history. The Skill does not require conversations to be uploaded to a third-party analysis service.
Before publishing a report or screenshot, review it for names, internal project details, credentials, private URLs, contact information, and anything unrelated to the intended assessment.