Skip to content
SkillScript-led1,758 tokens in use

Software Factory

Let coding agents ship fixes through reviewed pull requests, capped merges at a pinned commit, dated backups and a 24 hour rollback watch.

3,862 stars on the repository MIT Licenseericosiu/ai-marketing-skills · software-factory
npx skills add ericosiu/ai-marketing-skills -s software-factory
View source

Install command as written in the repository's README

Skill label

measured from its files

What installing Software Factory costs

30tokens idle
1,758tokens in use
4,155tokens, up to

Idle is what sits in context while the skill is installed. In use is its instructions, loaded when it runs. Up to adds the files its instructions link. Counts are approximate, made offline with cl100k_base.

Model dependence Script-led

Script-ledMixedModel-led

Most of this skill is scripts, so it behaves much the same whichever model runs it.

This describes the skill's structure, how much of it is scripts. It does not say how well it works. Scripts: scripts/automerge.py, scripts/blast_score.py, scripts/daily_card.py, scripts/factory_gh.sh, scripts/factory_ledger.py, scripts/factory_paths.py, scripts/github_app_token.sh, scripts/make_watch_config.py, scripts/not_live_sweep.py, scripts/portback.py, scripts/postrelease_watch.py, scripts/record_attempt.py, scripts/scrub_snapshot.py.

Touches

shelldetectednetworkdetectedwrites filesdetectedgitdetectedcredentialsdetected

Declared tags come from the skill's frontmatter; detected ones from its scripts, its code blocks and the commands it names in inline code. This is detected, not audited.

How it scores

Quality

How well the skill is written, counted from its own files: its frontmatter, its steps, examples, stated output and constraints, its size and whether every file it ships is one it names.How we score it →
Good
  • Frontmatter8/15
  • Structure19/25
  • Examples15/15
  • Output10/15
  • Size15/15
  • Resources10/15

Security

Starts at 100 and loses points for each thing in the skill's files that a reader should look at first.How we score it →
Needs attention
  • Runs shell commands−5

    does this on the machine, and the frontmatter does not say so

    SKILL.md, line 12
  • Writes files−8

    does this on the machine, and the frontmatter does not say so

    SKILL.md, line 74
  • File that was not read−15

    a file the skill ships that was not read, so what it does is unknown

    references/ci.yml.example
  • File that was not read−5

    a file the skill ships that was not read, so what it does is unknown

    references/gitignore.example
  • File that was not read−5

    a file the skill ships that was not read, so what it does is unknown

    references/gitleaks.toml.example
  • Runs git commands−5

    does this on the machine, and the frontmatter does not say so

    scripts/blast_score.py, line 57
  • Reaches the network−10

    does this on the machine, and the frontmatter does not say so

    scripts/blast_score.py, line 57
  • Reads credentials−15

    does this on the machine, and the frontmatter does not say so

    scripts/factory_gh.sh, line 44

Repository health

Pre-computed score based on development activity, maintenance, community, maturity, and trend momentum.How we score it →
Good
  • Development activity88
  • Maintenance44
  • Community56
  • Maturity28
  • Momentum40

Software Factory is a guarded loop for coding agents. Builder agents open pull requests with an acceptance test and stop. An operator merges at a pinned commit SHA through a GitHub App limited to contents and pull requests, under a daily cap that defaults to two.

After a merge, a port-back script updates the live copy only after a shadow test passes and a dated backup is made. A 24 hour watch can restore every file if a metric or error threshold is breached. A set of Python scripts and config files carry each step.

What You Get

  • A ten-step loop covering intake, blast scoring, builders, review, merge, port-back, watch and ledger
  • A blast-radius score for each diff that reviewers can raise and nothing can lower
  • Pinned-SHA squash merges through a GitHub App with a daily cap and optional blocked weekdays
  • Port-back with a shadow test, per-file dated backups and a 24 hour restore watch
  • Append-only ledgers, a not-live sweep and a daily card showing what is stuck

Common Use Cases

  • Letting coding agents ship small fixes while a human controls what merges
  • Putting a merge cap and rollback watch around agent-written changes to a live system
  • Adding a draft-only lane for repositories where a person always merges
  • Scrubbing and leak-scanning a repository before it joins the loop

What Makes It Unique

Most agent coding workflows stop at opening a pull request. This one also governs the merge and the release: merges are pinned to a SHA and capped, the live copy changes only after a shadow pass, and a restore path is armed afterward. The rules are explicit that agents never merge, never use admin flags and treat inbound text as data.

Under The Hood

The agent runs a set of Python and shell scripts from the repository checkout. They score diffs, call the GitHub REST API through a GitHub App using credentials stored in a folder you set, merge pull requests with a pinned SHA, copy files into a live directory with dated backup copies, run watchers on a schedule you start and write append-only ledger files. Reviewers are other coding agent command line tools. The SKILL.md preamble also runs two telemetry helper scripts that log usage locally and ask once about opt-in remote telemetry. Shadow tests stub sends and drop tokens from the environment, and the merge and port-back commands need an explicit execute flag.

Join founders building with open source

Opinionated takes, migration guides, cost-saving tips, and insights from the open source ecosystem.

Subscribe on SubstackJoin 750+ subscribers