lazycoder 0.1.0

Code review agent with senior-level judgement: interrogates every diff hunk against a fixed rubric and returns APPROVE / REQUEST_CHANGES / BLOCK

 lazycoder

A code review agent with senior-level judgement. It interrogates every changed block against a fixed rubric, runs the real checks, and returns a defensible verdict — APPROVE / REQUEST_CHANGES / BLOCK — before code is trusted or merged.

Code gets written fast. The bottleneck is trusting it. lazycoder is the reviewer that never gets tired, never skips a rule, and never self-reports green without running the checks.

Manual review vs lazycoder

	Manual review	lazycoder
Coverage	Whatever the reviewer remembers to look at	Every rule (R1–R17) evaluated, every time
Consistency	Varies by reviewer, mood, time of day	Same rubric, same policy, deterministic
Verdict	"LGTM" / gut feel	APPROVE / REQUEST_CHANGES / BLOCK from a severity policy
Evidence	Comments, sometimes	Every finding cites rule_id + exact file:line
Green claims	"tests pass" (trust me)	Real linter/typecheck/test output in a sandbox
Untrusted code	Reviewer may run it locally	Reviewed code is data, never executed outside the sandbox
Speed at scale	Slows down as diffs grow	Loops the rubric per block, unattended
Auditability	Lives in someone's head	Append-only decision log; any verdict is replayable

lazycoder does not replace the human — a person still confirms consequential decisions. It removes the parts humans are bad at: remembering all 17 rules, staying consistent across 200 files, and proving the checks actually ran.

Two structural facts, at a glance. These are not benchmarks — they are properties enforced by the schema, so they hold on every single review:

xychart-beta
    title "Rubric rules guaranteed evaluated per code block"
    x-axis ["manual review", "lazycoder"]
    y-axis "rules (of 17)" 0 --> 17
    bar [0, 17]

Manual review may cover all 17 — nothing guarantees it. lazycoder cannot emit a verdict until every rule has a recorded pass/fail (APPROVE is refused otherwise).

xychart-beta
    title "Findings that cite rule_id + exact file:line (%)"
    x-axis ["manual review", "lazycoder"]
    y-axis "% enforced" 0 --> 100
    bar [0, 100]

A human reviewer can cite evidence; the lazycoder domain model makes an uncited finding unrepresentable — pydantic rejects it before it exists.

Status

The full pipeline is live end to end — deterministic core plus the real model. A unified diff flows all the way to an aggregated verdict:

diff → parse_diff → CodeBlock[]
         └─ review_rubric(block, rubric)  # every rule, every block
              └─ RuleResult[] → from_rule_results → aggregate → verdict

The same flow runs in two modes, sharing every line of plumbing:

• Fake client (default, CI): deterministic, network-free. pytest -q proves the parser, aggregator, and verdict policy on every run.
• Real client (opt-in): AnthropicClient hits the live API. The first live run of eval E3 already passed — the model caught the SQL injection, flagged R7, and the pipeline derived BLOCK with zero parse failures.

Because the model was the last thing plugged in, any failure isolates to the prompt or the model — never to the plumbing, which is already proven. The response parser is hardened against real LLM output (code fences, surrounding prose, severity casing), and the reviewer prompt teaches the model the exact Finding schema with a literal example, so form errors die at the source.

Config-driven policy

Policy is declarative and lives in config/, not buried in code. Each file is one part of the setup — reviewable, diffable, swappable:

lazycoder/
├── config/
│   ├── harness.json              # project context, stack, hard rules, definition of done
│   ├── guardrails.json           # what the agent may / may not do; injection defense; limits
│   ├── setup.json                # runtime, deps + rationale, env vars, bootstrap
│   ├── working_loop.json         # specify → plan → execute → verify → decide
│   ├── task_loop.json            # orchestrator + review subagents, isolation, aggregation
│   ├── review_rules.json         # R1..R17 — the interrogation rubric (the core)
│   ├── production_readiness.json # the release gate
│   ├── evals.json                # known-flawed/clean cases that test the reviewer
│   └── observability.json        # append-only decision log, tracing, redaction
├── src/argus/                    # domain, config loader, reviewers, llm client
└── tests/                        # unit + integration + eval coverage

The rubric (R1..R17)

Code-level: data structure (R1), control flow (R2), inputs/outputs (R3), failure modes (R4), side effects (R5), dependencies (R6). Security: validation, secrets, injection (R7). Simplicity: simplest form (R8). System-level: state (R9), sync vs async (R10), monolith vs services (R11), invariant (R12). Plus maintainability, tests, and compatibility rules through R17.

Design decisions — the why

The interesting part of this project is not the review logic; it's the choices that make the review logic trustworthy.

• Deterministic core, model last. Everything that can be pure logic is pure logic, and the non-deterministic LLM is bolted on at the very end. This is a deliberate failure-isolation strategy: when a review goes wrong, the bug is in the prompt or the model, because the plumbing has tests proving it isn't there.

• Contracts make invalid state unrepresentable. The domain types are strict pydantic models with validators, not bags of fields. A passed rule cannot carry a finding; a failed one must. Every finding must cite its rule_id and an exact file:line. The verdict is a computed field over findings, never a value someone can set by hand. You cannot construct a lying ReviewReport.

• Normalize at the boundary, keep the core strict. Untrusted LLM text is cleaned up where it enters ("HIGH" → "high"), but the domain enum stays the single source of truth and never loosens. Leniency lives at the edge; the core does not bend.

• Debt is executable, not documented. The one known parser limitation is pinned by a strict xfail test, not a comment someone can ignore. The day the fix lands, that test flips to green and the suite tells you the debt is closed. Notes rot; tests don't.

• TDD throughout. Every behavior went RED before GREEN — including the garbage-input fixtures that hardened the parser.

• The eval is the product. config/evals.json is a set of known-flawed and known-clean cases whose job is to measure the reviewer itself. Wired as a CI gate, it closes the loop: a code reviewer that has its own reviewer, and knows whether it's still good every time it changes.

Develop

uv sync --extra dev
pre-commit install

pytest -q                       # deterministic suite — no network, no key
ruff check . && black --check .
mypy src

To run the live-API suite (opt-in, never part of pytest -q):

cp .env.example .env            # fill in ANTHROPIC_API_KEY — .env is gitignored
set -a; source .env; set +a
pytest -m integration

Roadmap

1. Multi-file / diff orchestration on top of review_rubric. ✓
2. Harden the response parser against real LLM output (fixtures). ✓
3. Wire config/evals.json as a regression gate on the fake client — a missed rule fails the gate. ✓
4. Wire the real Anthropic client behind the same LLMClient protocol, with an opt-in integration suite (pytest -m integration). First live run: the model caught eval E3's SQL injection (R7 → BLOCK). ✓
5. Run the full evals.json set against the live model and track the score over time — the eval stops measuring the plumbing and starts measuring the reviewer: does this prompt, on this model, still catch what it must?
6. Distribution: package for PyPI with a lazycoder console entry point, so users install with uvx lazycoder / pipx install lazycoder and review a diff with one command. A GitHub Action wrapping the same CLI comes after.