Skip to content

About

No description, website, or topics provided.

Resources

Code of conduct

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

 
 

Latest commit

 

History

89 Commits

Folders and files

Repository files navigation

PromptCite

Cross-agent /receipt for honest AI use disclosure on student work.

License: AGPL v3 CI CodeQL Release OSSF Scorecard

Install

macOS, Linux, WSL, Git Bash

curl -fsSL https://raw.githubusercontent.com/camadkins/promptcite/main/install.sh | bash

Windows (PowerShell 5.1+)

irm https://raw.githubusercontent.com/camadkins/promptcite/main/install.ps1 | iex

Installs into every supported AI agent on your machine. ~30 seconds, idempotent, safe to re-run. Per-agent commands and flags in INSTALL.md.

What it is

PromptCite installs into 26 AI coding/writing agents (Claude Code, Gemini CLI, Cursor, Windsurf, Codex, Copilot, Cline, Continue, Roo, Aider, Amazon Q, Kiro, Augment, Trae, JetBrains Junie, Goose, Replit, OpenHands, Qodo, Zed, opencode, Amp, Crush, GitHub Copilot CLI, Factory Droid, Jules) — plus any other AGENTS.md-aware agent automatically, and anything else via promptcite --print-rule — and adds a single command: /receipt. After AI-assisted work, a student runs /receipt, answers a handful of questions in under two minutes, and gets back three things to paste into their submission:

  • A formatted citation (MLA, APA, Chicago, IEEE, or Harvard).
  • A plain-language disclosure paragraph for the paper's header.
  • A structured JSON receipt for instructor review and archival.

One receipt covers one assignment, not one sitting. Run /receipt again a few days later and it adds that session to the existing receipt rather than replacing it — an assignment worked across a week ends up as one honest record.

Optionally, an AI-use ledger records your insertions locally so /receipt reminds you what you did instead of asking you to remember. Off by default, separately installed, stored outside your repo, and purged once used — it exists to make honest disclosure easier, never to report on you. If you run /receipt late, after your agent has compacted or cleared its context, the hook can also point /receipt at the agent's own local transcript so the earlier sessions aren't lost. It stores only where that file is, never what is in it, and nothing leaves your machine. See INSTALL.md.

Optionally, for courses where the point is that you own what you verify, PromptCite can mark AI-written code @ai-unverified and ask you what you checked before you keep it. Your answer goes in the comment, in your words, and promptcite-check lists anything you haven't got to yet. It's a to-do list you run on yourself, not a report anyone else reads.

Recording is per folder and off by default: promptcite on when you start an assignment, promptcite off when it's handed in, promptcite status to see where you are.

Different classes, different AI rules. Hand PromptCite a syllabus once with /receipt course add syllabus.pdf and it saves that class's AI-use rules, each one quoted word for word from the syllabus, to a course file on your machine. Bind a homework folder with promptcite course use csce-423, or just say "working on CSCE 423," and /receipt follows that class's rules while /receipt ask "can I use AI to check my proof?" answers from the syllabus. Where the syllabus says nothing, it tells you to ask your instructor rather than guessing. See INSTALL.md.

It is not an AI detector. It produces no originality scores. It is a transparency artifact for assignments where AI use is permitted and disclosure is required.

See it in action

Student finishes a policy essay in Claude Code. Fresh session.

> /receipt

Are you disclosing AI use from this current session, or previous?
> this session

A few quick details, answer all:
  1. course, instructor, assignment title
  2. your name or ID for this receipt
  3. citation style (MLA / APA / Chicago / IEEE / Harvard)
> ENGL 251, Dr. Martinez, Policy Analysis Essay, C. Hawkins, MLA

What did you use the AI for?
  brainstorm / outline / draft / edit / debug / explain / search
> brainstorm

  (1) one-sentence summary of what you asked the AI to brainstorm
  (2) did any AI-generated text appear verbatim?
> counterarguments to carbon tax policies / no

Anything to add for the revision statement?
> Used the list to structure my own outline, rewrote arguments in my own words.

══════ AI Use Receipt ══════

CITATION (MLA):
  "Counterarguments to carbon tax policies." prompt. *Claude*,
  Opus 4.7 version, Anthropic, 14 May 2026.

DISCLOSURE:
  I used Claude (Opus 4.7) on 2026-05-14 to brainstorm counterarguments
  to carbon tax policies for this assignment. No AI-generated text
  appears in the final submission. I used the list to structure my own
  outline and rewrote the arguments in my own words. This receipt was
  generated inside Claude itself, so the tool, model, and date fields
  above were agent-verified rather than student-reported.

JSON receipt saved as ai-receipt.json (content_hash dc84..4ed1).
Paste citation in your Works Cited. Paste disclosure in your paper's
header. Attach ai-receipt.json to your submission.

Five turns. Under two minutes. Receipt is yours to submit.

A real sample receipt is in examples/brainstorm-receipt.json. The full instructor-side reading guide is in docs/for-instructors.md.

Verifying a receipt

Each receipt includes a content_hash field (sha256 of the canonical other fields). Instructors can verify the receipt was not edited between generation and submission:

npx -y github:camadkins/promptcite promptcite-verify receipt.json

Exit 0 means the hash matches. Tamper-evident, not tamper-proof: the algorithm is public and a determined student can recompute the hash after editing. For most disclosures this is a meaningful speed bump. Full discussion in docs/for-instructors.md.

License

PromptCite is open source under AGPL v3, for everyone: students, instructors, schools, researchers, and anyone building on it. Use it, fork it, change it. If you run a modified version as a service, share your changes.

Ideas, issues, and PRs are all welcome. See CONTRIBUTING.md.

Docs

About

No description, website, or topics provided.

Resources

Code of conduct

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages