ForgeCode
Hacker ApprovedRanked 3 of 11 coding agent harnesses by recommenders
Free coding agent harness that people credit with reliable tool calling for small local models and strong Terminal Bench results; critics question its benchmark claims.
Free coding agent harness that people credit with reliable tool calling for small local models and strong Terminal Bench results; critics question its benchmark claims.
51 people gave an opinion
- recommend it
- 39
- criticize it
- 12
- approval share
- 76%
Recommended in 10 of 13 months
- November 2025: recommended
- December 2025: not recommended
- January 2026: recommended
- February 2026: recommended
- March 2026: recommended
- April 2026: recommended
- May 2026: recommended
- June 2026: recommended
- July 2026: recommended
- August 2026: recommended
- September 2026: not recommended
- October 2026: recommended
- November 2026: not recommended (month in progress)
Why it carries the badge
- 39 people recommend itNeeds 10 or more, met.
- 3.3 recommenders for every criticNeeds 2 to 1 or better, met.
- Recommended in 10 of 13 monthsNeeds 4 or more, met.
The same three rules apply to every product, and the badge can’t be bought. How the badge works
What people say about it
The reasons people give, with the number of comments that give each one. Open a reason to read what they said.
Why people recommend it
Each mark is one comment
Tool calling for small models
“Forge just increases mechanical reliability of tool-calling, it shouldn't disrupt your workflow-level layers much.”
Comment from 20 May 2026 on Hacker News “Forge provides a more hardened suite of guardrails that are effectively necessary for small models.”
Comment from 20 May 2026 on Hacker News “It can bring 8b models up to Opus-like performance at tool calling.”
Comment from 14 Jun 2026 in “RTX 5080 and RTX 3090 Setup: 80 Tok/s on Qwen 3.6 27B Q8” Strong Terminal Bench results
“ForgeCode appears to be pretty good (currently, #1 on Terminal Bench 2.0”
Comment from 23 Apr 2026 in “Ask HN: Open-Source Coding Model and Harness at Claude Sonnet / Opus Level Perf?” “has better results than Open Code, at least according to Terminal Bench”
Comment from 4 May 2026 in “DeepClaude – Claude Code agent loop with DeepSeek V4 Pro, 17x cheaper” Preferred over Claude Code
“I really liked forgecode the best, and it feels better than claude code”
Comment from 19 May 2026 in “Cursor Introduces Composer 2.5” “Firstly, forgecode is a much better harness than Cloude code (IMHO).”
Comment from 12 Jun 2026 in “Kimi K2.7-Code: open-source coding model with better token efficiency” Structured planning and step-by-step execution
“Forge brings rigorous, adversarial architectural planning and reliable step-by-step execution to AI-assisted development.”
Comment from 24 Feb 2026 in “Adversarial multi-agent planning and deterministic pipeline execution for agents” “keeps a decision log and a dependency-aware task graph so changes remain explainable and resumable”
Comment from 9 Mar 2026 in “Show HN: Forge – decision logs and task graphs for AI coding assistants”
What people warn about
Each mark is one comment
Accused of gaming benchmarks
“I cannot believe they didn't do something to influence the benchmarks”
Comment from 23 Apr 2026 in “Introducing GPT-5.5” “ForgeCode for instance has been cheating in the past”
Comment from 4 May 2026 in “DeepClaude – Claude Code agent loop with DeepSeek V4 Pro, 17x cheaper” Stalls when model replies with text
“But at the end of the day, if the model keeps responding with text, there's nothing forge can do.”
Comment from 20 May 2026 on Hacker News “if the model keeps responding with text, there's nothing forge can do”
Comment from 20 May 2026 on Hacker News
Other coding agent harnesses people recommend
The four most recommended other coding agent harnesses, with the same counts for each, so you can compare like for like. The tick on each bar marks the 2 to 1 line the badge needs.
| Product | Recommend | Criticize | Approval | Months | Owned | Badge |
|---|---|---|---|---|---|---|
| Pi coding agent | 892 | 198 | 82% | 12 | 3 mo51 owners | Approved |
| omp (oh-my-pi) | 80 | 21 | 79% | 6 | — | Approved |
| ForgeCodeThis page | 39 | 12 | 76% | 10 | — | Approved |
| Reasonix | 26 | 19 | 58% | 6 | — | — |
| DSH | 11 | 1 | 92% | 3 | — | — |
What people said
40 of the 75 comments that mention it, newest first. Each links to the original comment.
“when are you going to support SHA256? I have SHA256 repo and I don't want to rewrite all history”
“allows me to customize the keyboard shortcuts for tiling and managing OS windows”
“only the featurs that you'd like to edit, logo, qr is changed in 10+ creatives at the same time”
“the March package manager ForgePM will reject packages that falsify their cap manifest”
“I'm currently using Forge but sadly it's becoming unmaintained”
“Using lightweight formal methods to verify models that the AI later builds.”
“the harness I use with a built-in engineering philosophy for everyday work”
“I have been inpressed with what it can do given all the limitations of local models”
“For long context 150k+) it start degrading and forgetting.”
“It can bring 8b models up to Opus-like performance at tool calling.”
“Firstly, forgecode is a much better harness than Cloude code (IMHO).”
“it is designed _not_ as an SDK but rather as a standalone, full-featured harness with an API”
“The core design decision that supports everything is having a sort of multi-tool that does a lot for the language called forge.”
“I'm guessing it's going to be an absolute banger of a month for forge”
“It's how I've been using coding agents myself for the past few months”
“Forge just increases mechanical reliability of tool-calling, it shouldn't disrupt your workflow-level layers much.”
“getting things stable enough for production use at reduced costs”
“if the model keeps responding with text, there's nothing forge can do”
“But at the end of the day, if the model keeps responding with text, there's nothing forge can do.”
“Forge provides a more hardened suite of guardrails that are effectively necessary for small models.”
“I've personally tried qwen3.6-35b-a3b, qwen3.5-35b-a3b, and qwen3.5-27b with forge (agentic coding harness built on forge workflowrunner) and it works great.”
“Forge will help make sure that the API calls to hotel booking sites parse correctly or gracefully retry.”
“sending the model a friendly and gracefully handled error message works wonders”
“But obviously as a drop-in replacement in existing systems then proxy is the way to go.”
“I really liked forgecode the best, and it feels better than claude code”
“ForgeCode for instance has been cheating in the past”
“has better results than Open Code, at least according to Terminal Bench”
“forge addresses this issue by enabling Claude Code to persistently retain contextual information.”
“I cannot believe they didn't do something to influence the benchmarks”
“ForgeCode appears to be pretty good (currently, #1 on Terminal Bench 2.0”
“Forge Optimizer — free tool for model routing, prompt compression, and AI cost projection”
“With a better harness like Forge Code, I have better results”
“It's one of the best ways to play the game: single-player, offline, and unofficial.”
“I now use Forge to plan css too because I don't want to show broken css states”
“keeps a decision log and a dependency-aware task graph so changes remain explainable and resumable”
“Forge brings rigorous, adversarial architectural planning and reliable step-by-step execution to AI-assisted development.”
“It flattens arrays and objects 5+ levels deep into relational tables with proper keys.”
“AI can execute them but cannot modify them, and every run is hash-verified and logged.”
“reducing a 6–12 hour workflow to 30–60 minutes on real feature builds”
“You'd be surprised how far you can get with using string identifiers for dependencies”
Counts come from 1,795,640 Hacker News comments, November 2025 to November 2026. How we count. Report a mistake.