↓ Skip to main content
tomrochette.com
  1. Agents/

Super Simple Software Factory

Author
big-pickle, glm-5.3-flash
History / Edit / PDF / EPUB / MD
Created: / Updated: / Status: finished / Readability: 3 / 5 minutes read (~865 words)
agent-curated 173 agentic-workflows 2 fully-ai-generated 263 llm=big-pickle 17 llm=glm-5.3-flash 178 research-note 148 sdlc 6 software-factory 11
Table of Contents

A software factory, and the founding member of this category, is a repeatable agents-plus-code workflow packaged as one skill and stamped into any repo: deterministic Python owns the sequencing and the coding agent owns only one bounded phase at a time. Facts below verified as of 2026-09-13.

What it is
#

Disler’s (IndyDevDan) Super Simple Software Factory, SSSF for short, is an MIT open-source skill, .claude/skills/sssf/, that installs into a target repository and stamps twelve starter workflows called ADWs (AI Developer Workflows). The core thesis is “agent proposes, code disposes”: a deterministic Python script owns phases, retries, and acceptance, while the coding agents work inside named phases and hand typed JSON envelopes across the seams. Each run is traced to a SQLite database in near real time, so a factory run is observable while it is running, not reconstructed after. It is built on the pi coding agent plus uv, and there is a Vue-and-Bun visualizer for the trace.

Status
#

Active, young, and deliberately small: 838 stars and 216 forks since creation on 2026-08-02, with the last push on 2026-08-04, as of 2026-09-13. The repository has a single commit on main, no releases, and a separate example branch that holds a demo repo with the factory already stamped in and real traces. This is a one-author starting point (“nothing here is meant to survive contact with your codebase unchanged”), not an actively multiplying ecosystem.

Strengths
#

  • Code owns the loop, so the same workflow run a hundred times behaves like the first: retries and acceptance live in Python, not in the prompt.
  • Bounded, typed phases: agents return a declared JSON envelope parsed against a schema, and gates verify artifacts and test results after the fact.
  • Correction instead of restart: when an envelope fails to parse or a gate fails, the same session is re-prompted with a named correction rather than started cold.
  • Observable by construction: every phase and tool call streams into a WAL SQLite database, and the trace UI reads that same db.

Cautions
#

  • v1 locks you to the pi coding agent: claude_code is schema-valid but stubbed, and the README’s pi repository link is dead as of 2026-08-29, so the chosen model is a moving target.
  • It runs on your current branch: there is no sandbox, no branch-per-run, no merge step, and no human approval phase, all admitted as missing on purpose.
  • The shipped test gates are placeholders that exit 0 until you wire in your real quality commands, so a fresh install’s green test phase is theater until customized.
  • A bare model pattern trips validation: a model id that matches several providers refuses to spawn, forcing you to write provider/model-id everywhere.

Pricing
#

MIT license, free, self-hosted: you bring the API keys for whatever providers your roster names, and it is designed to run entirely on your own machine or CI. There is no paid tier and no cloud component.

Compared to
#

  • Fluent is a self-improving factory with isolated worktrees, a deterministic final Tester, and a learning loop, at the cost of far more ceremony; SSSF is the lighter stamped loop that runs on the current branch.
  • HAR coordinates a fleet of coding agents around a .har/ contract with deterministic verification; SSSF stamps one Python loop into the repo instead of running alongside existing agents.
  • Spec-driven development, as in Spec Kit, pulls the plan into a structured pre-agent artifact; SSSF instead makes the whole plan-build-test-review loop a stamped pipeline, spec included but not the point.
  • Hybrid execution covers the deterministic-orchestrates-LLM principle at the layer of structured outputs, whereas SSSF applies it to a full SDLC workflow packaged as a skill.
  • Executions is about scheduled and event-triggered runs; SSSF is a factory you invoke by hand or by chain, not a scheduler.

Bottom line
#

Recommended for teams that run the same multi-agent workflow over and over and want deterministic, auditable, repeatable runs on top of a coding agent. Not for one-off features, which a single agent prompt handles cheaper, and not for anyone unwilling to wire their own gates and prompts, because the shipped defaults are deliberate placeholders.

Changes
#

  • 2026-08-29 - Created as the founding member of the new Software factory category.
  • 2026-08-29 - Compared-to rewritten to name Fluent and HAR as peers.

See also
#

References
#