Setlist emerging

Core approach

A human directs as architect and reviewer while a Planner writes one small spec with explicit out-of-scope boundaries and exhaustive acceptance criteria, then a Builder implements it. Two QA passes and a completed Closing report gate closure; stamped Git hooks check commits and merges and run a push-time audit of trunk history, while the Claude Code session hooks are advisory.

Workflow

1. /setlist:new — bootstrap an empty directory through an interview, approved foundational decisions, and a two-phase project stamp
2. /scaffold — use the generated project-local skill to wire the test harness, record the gate command, create the first commit, and arm the Git hooks
3. Plan one spec at a time — the Planner re-grounds from repository artifacts, writes explicit scope and acceptance criteria, and parks ambiguities instead of improvising
4. /setlist:checkpoint — open the spec branch, survey and commit spec-scoped work, run the close gate, and merge with --no-ff
5. Build and verify — the Builder implements in small commits, then records automated criterion-by-criterion QA and a human spot-check in the Closing report
6. /setlist:validate and /setlist:gate — check instance health and govern transitions between roadmap stages
7. /setlist:journal — capture a substantive session's raw findings and surprises in the repository
8. /setlist:retrofit or /setlist:upgrade — adopt Setlist around an existing codebase or migrate an earlier framework edition

Supported tools

• Claude Code

Notes

Setlist makes the repository the durable memory for an AI-directed project: specs, a bounded status file, decision records, journals, and a living architecture diagram survive session resets, while a Planner/Builder split maps onto Claude Code’s plan and execution modes. Its published operating loop is intentionally serial: plan one spec, branch, build, run two QA passes, complete the Closing report, close, then repeat. The framework document defines the spec contract in detail, including explicit out-of-scope items, acceptance criteria, QA verdicts, and the required Closing report. See the versioned README’s principles and operating loop and the v1.11 spec template.

The strongest differentiator is mechanical enforcement. A generated instance includes tracked Git hooks for early commit and merge checks plus a pre-push audit that examines trunk history for code that did not arrive through a closed spec. Setlist is careful not to overstate that mechanism: its documentation calls it a discipline control for cooperating use, lists known ways determined committers can evade local enforcement, and directs security-sensitive users to forge-side branch protection and required checks. The same documentation also records a current harness limitation: advisory session-hook reasons are dropped on Claude Code’s allow path, leaving Git-hook refusals as the feedback users actually see. See the trunk-audit and limitations documentation and the stamped Git-hook templates.

The project reports both cold end-to-end dogfood runs and hostile subprocess tests before release, with an automated hook suite running on Linux and macOS. Its rapid release history reached plugin v2.3.0 and framework edition v1.11 in August 2026, which supports an emerging maturity rating: the mechanics are substantial and actively tested, while the published limitations remain extensive and the interface is still changing. See the testing account, the test suite, and the first-party release history.

Added Aug 31, 2026 · Assessment last reviewed Aug 31, 2026 · How we track