Repository navigation
3.6.0
Dedicated agents for repeated tasks
At the close of each work, Playbook now records the tasks the agent carried out, and proposes a dedicated agent, on a model that fits the task, for a task that repeats (work task-agents, #44).
-
Tasks recorded. At the close,
prepare-projectreads the git log since the previous work's close, so it also sees the previous release. It records each task it would carry out again, such asreleaseorclose-stage, indocs/wiki/tasks.mdwithtasks.sh, a new script ofplaybook-agents. -
A proposal at 3 works. When a task is held in 3 distinct works, the agent judges its weight:
- mechanical steps go to
haikuwith effortlow; - writing text or judging a small change goes to
sonnetwith effortmedium; - deciding, designing, a push, a merge, or a release stays in the main session.
It proposes one agent at a time and always asks you.
- mechanical steps go to
-
Your answer. On a yes, it writes
.claude/agents/<task>.mdin the project, with only the steps it may run and a line naming those left to you. On a no, the task is marked declined and is not proposed again. -
Six general rules learned from repeated review findings now ship in
evidence-rules.md: three new, and three raised to level 3.
Nothing to migrate: the wiki format stays 5. The step runs only when playbook-agents is installed; docs/wiki/tasks.md is created on the first close.