Examples

​

Run one

1. Set the project up as Installation describes.
2. Copy the file and the folder beside it, its brief and sample inputs, into the directory the work belongs in.
3. Sign in to the coding agents the file names, Claude Code and Codex in most of them.
4. Run it with npx tsx <file>.ts.

examples/
  feature-team.ts
  forge-helper.ts
  teams/
    feature-delivery.ts
    briefs/triple.md
  use-cases/
    engineering/
      backlog-groom-then-rank.ts
      briefs/backlog.md
      backlog/raw.md
    business/
      contract-playbook.ts
      briefs/playbook.md
      contracts/msa.md
    other/
      translate-reflect.ts
      briefs/translation.md
      glossary/en-fr.md
      source/article.md

​

Product development

• Feature delivery. Seven steps from a ticket to a change that lands: a build held to green tests, a goal check, three reviewers whose notes become one list, a judge and a person’s approval. The same team can work through a backlog.
• Shipping through GitHub. After the review passes: push the branch, open one pull request, pass a strict gate on the exact head revision, squash the merge.
• Backlog grooming. Raw tickets become stories with acceptance checks, a second model reads them back, and the product owner ranks.
• Shape Up cycle. One cycle composed of workflows: requests shaped into pitches, a betting table of three seats each checked by Jev, a person’s bets, a build inside the appetite, and a cool-down, with the hill read from the record.

​

Finance

• Market session advice. A session host reads recorded observations, a typed assessment routes each, an uncertain one is researched, and a person is asked where action would be considered.
• Invoice chase. A command reads the ledger, a model drafts a chaser per overdue invoice, the mailer sends the plain ones, and a disputed one stops on a question for a person with the draft beside it.

​

Business operations

• Contract review. Every clause mapped to the playbook rule it meets or breaks, redlines in the playbook’s wording, and a lawyer decides what is sent.
• Draft, then send. A model drafts one note per warm contact, a person decides on each, and the record keeps what was sent, what was refused and why, and what is still waiting.
• Triage with escalation. A typed classification routes each ticket, every ticket gets a drafted reply, the routine ones are answered, and the rest stop on a question for a person with the draft beside them.
• Shortlist. A script applies the filters and records each exclusion’s rule, two rankers from different model families rank the rest against the brief, a judge lands the better ranking, and a person decides who is shortlisted.
• Approve the payload. A person’s yes is recorded against the digest of the request’s bytes; the same bytes are sent, and a payload that differs by one field asks again.
• Handoff that resumes. A run stops after its first stage, the next worker is given only the record, and it carries on from the exact state; the finished stage never runs again.

​

Research and knowledge

• Literature watch. A model summarises the supplied papers, a person decides which summaries enter the lab’s memory, and a question is answered from the kept notes alone.
• Vault curator. A model proposes where each inbox note belongs and a cleaned version, a person decides what is filed and where, and a question is answered by searching the vault and citing the passages.

​

Writing and content

• Writer, grader, judge. One model drafts a post, a grader from another family returns findings against the house style, a judge says when another round stops being worth it, and an editor decides whether it is published.
• Translate and reflect. An article translated against a glossary, read back by an editor on another model, and a person decides the nuance.

​

Next steps