menu
run / passed · @codedology/code-forge

Run coding agents in verified blocks.

Every change is planned, reviewed in a fresh session (a second model if enabled), proven by gates, and recorded with its cost and time.

npm install -g @codedology/code-forge
code-forge init
Open-source (MIT) CLI · Claude Code, Codex, Grok … · Cost & time ledger
code-forge ledger · 2026-10-20
recorded
01 / LEDGER REPORT LINES COST & TIME
r-20261020-055623-e76b74 | 1h 18m 46s | $11.9205 | 3 closed | 1 stopped
B147 | $3.2928 | 39m 11s | stopped

Why stopped matters: cost, time and outcome are recorded even when a block stops.

02 / GATE OUTPUT FROM A REAL BLOCK PROVEN
test pass 3544, fail 0 · types ok · lint skipped · format skipped
03 / REVIEW RESULT severity: warning

"The evidence pattern blocks a drive path only as a prefix, so ./C:/Windows still validates."

severity: warning → closed by one small patch
Real ledger, gate and review excerpts. Shown together; not from the same block.
01 SOLO / AGENCY FOCUS
SOLO DEVELOPERS ● SELECTED

“Know what every task cost, and that it passed.”

For solo developers who run several coding agents and want review, proof and cost control on every block.

Small run example: 1 block · 13m 19s · $1.72 Example from code-forge’s own ledger; not a cost estimate.
SMALL AGENCIES ● SELECTED

“Price agent-built work from real numbers.”

For small agencies that build client work with agents and must price it from real cost and time recorded per block.

Overnight run example: 3 blocks · 1h 18m 46s · $11.92 Example from code-forge’s own ledger; your work will have its own cost and time.
02 HOW IT WORKS · FIVE-STEP BLOCK PIPELINE

Plan → Code → Review (fresh session) → Gates → Ledger.

STEP 01 · PLAN

Plan

Work is cut into blocks and each block gets a plan before an agent writes code.

BLOCK STRUCTURE BEFORE CODING
Work is cut into blocks (e.g. B147).
Each block gets a plan before an agent writes code.
STEP 02 · CODE

Code

One coding agent executes the planned block while code-forge tracks cost and time.

CODING AGENTS COST & TIME
Claude Code · Codex · Grok …
One agent codes the block; cost and time are recorded.
STEP 03 · FRESH SESSION

Review (fresh session)

A fresh session reviews the change and reports findings, so no session grades its own work. If you enable it, a second model reviews too.

REAL REVIEW FINDING WARNING → FIXED
"The evidence pattern blocks a drive path only as a prefix, so ./C:/Windows still validates."
severity: warning → closed by one small patch
STEP 04 · GATES

Gates

Every block is proven by gates. Configured gates must pass or be recorded as skipped before a block closes.

REAL BLOCK GATE OUTPUT PROVEN
test pass 3544, fail 0 · types ok · lint skipped · format skipped
test pass 3544, fail 0
lint skipped
types ok
format skipped
STEP 05 · LEDGER

Ledger

Each block is recorded in the ledger with its cost and time. Cost, time and outcome are recorded even when a block stops.

LEDGER REPORT LINES 2026-10-20
r-20261020-055623-e76b74 | 1h 18m 46s | $11.9205 | 3 closed | 1 stopped
B147 | $3.2928 | 39m 11s | stopped
03 PROOF FROM THE LEDGER

code-forge is built with code-forge

Source: code-forge’s own ledger · 2026-11-06
RECORDED RUNS
28 runs
In code-forge’s own ledger
VERIFIED BLOCKS
179 blocks closed
Planned, reviewed & proven by gates
RECORDED SPEND
$576.77
Recorded across 28 runs
ONE OVERNIGHT RUN r-20261020-055623-e76b74
BLOCKS 3 blocks
TIME 1h 18m 46s
COST $11.92
A SMALL RUN SINGLE BLOCK
BLOCKS 1 block
TIME 13m 19s
COST $1.72
Examples, not cost estimates: these numbers are examples from code-forge’s own ledger. Your own work will have its own cost and time.
04 AGENTS & KEYS

Works with your agents and your own keys.

You keep your own API keys. One agent codes, a different model reviews.

Claude Code · Codex · Grok …

One agent codes, a different model reviews. You keep your own API keys.

05 FAQ · 3 QUESTIONS

Questions & facts.

What is a block? [01]

Each block is a planned unit of work: planned first, coded by one agent, reviewed in a fresh session (a second model if enabled), proven by gates (the checks you configure), and recorded in the ledger with its cost and time—even when a block stops.

Which agents can I use? [02]

Claude Code, Codex, Grok and more, with your own keys. You keep your own API keys: one agent codes the block and a different model reviews it.

How does it help with client pricing? [03]

The ledger records cost and time per block, so you can price the next job from facts. The run numbers on this page are examples from code-forge’s own ledger; your own work will have its own cost and time.

INSTALL & INITIALIZE

Every change is planned, reviewed in a fresh session (a second model if enabled), proven by gates, and recorded with its cost and time.

code-forge is an open-source (MIT) command-line tool published as the npm package @codedology/code-forge. Install globally and initialize in your repository:

npm install -g @codedology/code-forge
code-forge init