PRismAI Pre-Review Assistant
Deterministic Grounding · Three Parallel Specialists

Checks your pull request before human review.
Then proves readiness.

PRism-AI coordinates Code Review, Testing, and Documentation specialists in parallel. It executes your project's real test suite and only says READY FOR HUMAN REVIEW when deterministic rules agree.

Run anywhere: pip install -e . && prismai
Install & Quick Start Explore Repository
438 / 438 Tests Passing
< 30s Parallel Review
0 AI Calls Re-check on Unchanged Files

Interactive PR Inspection

⚙ code-review 12s ✓
✓ testing 34s ✓
📖 documentation 25s ✓

Running 3 specialists across code, pytest suite, and OpenAPI surface...

ATTENTION REQUIRED
×

Traditional CI Only Checks Yesterday's Assumptions

If you introduce a new feature with an unhandled None crash or leave API documentation stale, existing unit tests still pass 100%. Reviewers waste hours on mechanical checks that slips right through CI.

✓

PRism-AI Checks the Full Quality Spectrum Before Review

It pairs real deterministic test execution with three AI specialists to catch implementation bugs, untested behavior gaps, and documentation rot in 30 seconds.

Three Specialists. One Deterministic Gate.

Ordinary code handles Git, test execution, schema validation, and policy. AI models provide deep reasoning over structured evidence.

Stage 1

Context Analyzer

Creates an immutable Git snapshot, extracts AST import graphs, maps callers and callees, and scrubs sensitive credentials.

  • ✓ Resolves Git diff & hunks
  • ✓ AST caller/callee graphing
  • ✓ Excludes secrets & binaries
Stage 2

3 Parallel Specialists

Three independent agents inspect the exact same immutable snapshot simultaneously across three distinct quality domains.

Code Review: Nulls, logic, validation
Testing: Pytest runner & test gaps
Docs: OpenAPI vs api.md sync
Stage 3

Result Aggregator

Deterministic Python validates JSON schemas against schemas/*.json. The AI model never decides whether code is ready.

  • ✓ Strict schema validation
  • ✓ 1 repair round on bad JSON
  • ✓ Deduplication with provenance
Stage 4

Delta Engine

When you fix bugs and re-run with --previous last, it diffs findings to report exactly what is Resolved, Persistent, or New.

  • ✓ Two-pass stateful comparison
  • ✓ Prevents regression blindspots
  • ✓ Clear audit-ready delta table

Engineered for Real-World Codebases

Built with production safety, multi-provider flexibility, and developer velocity at its core.

💬

Talk to It: prismai chat

Ask plain-English questions in your terminal: "Which parts are riskiest?", "Explain the bug in shipping.py", or type /check for a formal review.

🧪

Deterministic Pytest Execution

We don't guess whether tests pass. The testing agent runs real pytest, captures exit codes and JUnit XML logs, and surfaces real stack traces.

⚡

Zero-Cost AST Memory

Scales to 100+ files. Unchanged modules are remembered in an AST fingerprint cache, allowing instant re-checks in ~1 second with 0 external AI calls.

📂

Works Anywhere (Git or No Git)

Run against Git branches, raw folders (prismai check --scan .), before/after directory snapshots, or standalone .patch diff files.

🔄

Continuous Watch Mode

Keep prismai watch . running in a terminal. Every time you press Ctrl+S, it quietly checks affected files and updates PRISM-REPORT.md.

🩺

Self-Diagnosing: prismai doctor

One command checks your Python environment, dependencies, terminal encoding, and tests live connectivity to your AI provider.

Install Once. Call from Anywhere.

Just like git or docker, PRism-AI installs as a global terminal command.

1

Install the CLI

Clone the repository and install the prismai command in editable mode:

git clone https://github.com/Hackathon-Repo-Org/PRism-AI-PROD-.git
cd PRism-AI-PROD-
pip install -r requirements.txt
pip install -e .

✓ Installs prismai into your PATH on Windows, macOS, and Linux.

2

Connect Any AI Provider

Works with DeepSeek, OpenAI, Gemini, Claude, or local Ollama:

# Option A: Interactive setup wizard
prismai setup

# Option B: Set environment variable
$env:DEEPSEEK_API_KEY = "sk-..."  # or OPENAI, GEMINI

✓ Your key is saved locally in .prism.env and is never committed.

3

Verify & Use

Run doctor to ensure all systems are green, then point it at any folder:

# Check setup & AI connectivity
prismai doctor

# Run review or chat with codebase
prismai check
prismai chat

✓ Outputs instant findings and writes PRISM-REPORT.md.

Command Reference

A unified interface tailored for both interactive coding and automated CI gates.

Interactive pair programming with full context awareness of your repository.

PS C:\my-project> prismai

PRism-AI - Hi! I'm your pre-review code checker.
  Project  my-shop-app · 117 files · 726 KB
  AI       DeepSeek · deepseek-chat

you › which parts of the code are the riskiest?
  • [HIGH] Security: shopmart/reports/sales.py:7 (SQL injection)
  • [HIGH] Security: shopmart/users/auth.py:5 (hardcoded credentials)
  • [HIGH] Null: shopmart/orders/shipping.py:7 (crashes on empty address)

you › /check
  → Running 3 specialists in parallel... Writes PRISM-REPORT.md

Deterministic pre-review gate. Checks commits, branches, or raw folders.

PS C:\my-project> prismai check --previous last

PRism-AI  HEAD vs main  [DeepSeek: deepseek-chat]
Run run-002: 3 changed file(s). Running specialists in parallel...

  code-review    completed  12s
  testing        completed  34s
  documentation  completed  25s
  tests: 22 passed, 0 failed, exit code 0

Since run-001:
  ✓ Resolved: 9  ·  ↻ Persistent: 0  ·  + New: 0

RESULT: READY FOR HUMAN REVIEW
Report written to: PRISM-REPORT.md

Continuous background daemon that updates your report every time you save.

PS C:\my-project> prismai watch .

Watching C:\my-project
Each check writes PRISM-REPORT.md there. Press Ctrl+C to stop.

[12:15:02] change detected in service.py... checking...
  • code-review: 1 issue found (bulk_discount off-by-one)
  • testing: 1 missing test identified
[12:15:35] change detected in service.py... checking...
  • all checks passed → READY FOR HUMAN REVIEW

Waiting for changes...

Self-healing environment diagnostics with zero external dependencies.

PS C:\my-project> prismai doctor

PRism-AI doctor
  [OK  ] Python 3.12.10
  [OK  ] package jsonschema
  [OK  ] package pytest
  [OK  ] git 2.44.0
  [OK  ] AI provider - api.deepseek.com - deepseek-flash
  [OK  ] AI connection - the model replied 'OK'
  [OK  ] prismai command - C:\Python312\Scripts\prismai.exe

All good - try:  prismai chat

What Each Run Leaves Behind

Every PRism-AI run generates an immutable, inspectable run folder. No hidden state, no mystery verdicts.

runs/<runId>/
Isolated run directory with timestamp & candidate commit hash
report.md
Actionable human-readable report with exact file, line, and remediation
report.json
Machine-readable JSON schema validated against schemas/report.schema.json
delta.md
Stateful comparison with previous run (Resolved vs. Persistent)
context.json
Shared immutable snapshot data, AST import graphs, and exclusions
logs/pytest.log
Raw execution logs and JUnit XML from the project's actual test run

What "READY FOR HUMAN REVIEW" Means

Review readiness is a scoped, verifiable state — never blind approval to merge or deploy without human judgment.

  • All 3 Specialist Checks Completed: Code Review, Testing, and Documentation agents all executed within timeout.
  • Deterministic Tests Passed: The project's build and test commands exited with code 0.
  • Zero Policy Blockers: No unhandled exceptions, SQL injection, null dereferences, or contract drifts remain open.
  • Full Provenance Preserved: Every finding is anchored to an exact file path, line number, and verifiable evidence.
✓ READY FOR HUMAN REVIEW

Deterministic Policy Gate Satisfied

Run: run-20260927-7bf8c12 · Snapshot: 7bf8c12

Code Review 0 Blockers
Test Runner 22 Passed (Exit 0)
Documentation OpenAPI In Sync
Delta: 9 Resolved · 0 Persistent