Cultivate

Epic E1

Coordinated coding agents

Agents can take approved engineering tasks, work in isolated branches and expose their progress through issues and pull requests.

Purpose

A human can split implementation work between agents without losing track of ownership, files, tests or review state.

Current status

Working for Slop Filter and evidenced in Delegate instructions. The coordination record is strong; product decision quality is still mostly human-led.

Milestones

Working

Issue, branch and PR workflow is visible

Every approved implementation starts from an issue, uses a named branch and records acceptance evidence in a pull request.

Tested

Worktrees are mandatory for concurrent work

Agents use isolated worktrees and record the path, base commit and expected files before editing tracked source.

Depends on: E1-M1

Evidence

Verified

LinkedIn Slop Filter main at 8984c80

Local main and GitHub main matched after cleanup. README, AGENTS.md and recent pull requests were inspected.

Open source
Verified

PR #118: mandatory isolated worktrees

Merged on 2026-10-11 and updated AGENTS.md to require one worktree per issue branch.

Open source
Partial

Delegate AGENTS.md and agent work guide

Private repository instructions were inspected for coordination patterns. The source link and private implementation detail are not published.

Risks and open questions

Risks

  • Agents may comply with written instructions unevenly.
  • Logical conflicts can still occur even when worktrees isolate files.
  • Private context can drift away from the public issue record.

Open questions

  • Which claims should be machine-validated before an agent starts work?
  • How much handoff detail is enough for a replacement agent to resume safely?

Linked Wardley components

  • Agent harnesses
  • Model APIs
  • Coding models
  • Git
  • Worktrees
  • GitHub coordination

Next steps

  • Keep issue claims, PR evidence and worktree cleanup visible in every new repository.