using-t9x
Work inside a t9x workspace: record tasks, runs, and notes under .agents/ instead of scattering provisional material through the project. Use whenever a repository contains a .agents/ directory, and whenever you are about to write a TODO list, progress log, experiment record, or scratch script.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-16
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- using-t9x
- description
- Work inside a t9x workspace: record tasks, runs, and notes under .agents/ instead of scattering provisional material through the project. Use whenever a repository contains a .agents/ directory, and whenever you are about to write a TODO list, progress log, experiment record, or scratch script.
03 SKILL.md 正文
Using t9x
t9x separates the human-curated project from provisional agent work. Put intentions, attempts, interpretations, and scratch code under .agents/. Touch the human workspace only when explicitly asked or through an import or promotion command.
The workspace contains five kinds of material:
| Kind | Question | Location |
|---|---|---|
| Task | What should be done? | .agents/tasks/ |
| Run | What did we try? | .agents/runs/ |
| Note | What do we currently think? | .agents/notes/ |
| Script | What code did the agent write? | .agents/scripts/ |
| Skill | What reusable method exists? | .agents/skills/ |
Files are the interface. You may read and edit them with ordinary file tools. IDs such as qx3 are canonical and unique across .agents/; filenames may change.
Working loop
t9x ready
t9x show qx3
t9x run new qx3
# Record work in the run file as it happens.
t9x run finish f2m --outcome inconclusive
t9x note new 'What we learned' --related qx3 f2m
t9x close qx3
Use one run per attempt or session. A note is for an interpretation that outlives the run. Put scratch code in .agents/scripts/ and reference it from the run. Record newly discovered work with t9x task new '...'.
State transitions
Never edit status: by hand. Use the semantic commands:
t9x close qx3
t9x wontdo qx3
t9x reopen qx3
t9x block qx3 1v2
t9x unblock qx3
t9x relate qx3 f2m
Do not delete completed or rejected tasks. They are project history. Preserve all unknown YAML front matter fields when editing objects directly, and never change an object's id:.
Moving notes across the workspace boundary
Import an existing Markdown document into the agent workspace:
t9x note import docs/identification.md --title 'Identification' --move
Promote accepted agent material into the human workspace:
t9x promote .agents/notes/2026-08-27-identification.md docs/identification.md
Use --move or promotion only when the user wants the source removed. Both commands preserve the source if the destination cannot be committed.
04 同仓库更多 Skills
using-t9x
Work inside a t9x workspace: record tasks, runs, and notes under .agents/ instead of scattering provisional material through the project. Use whenever a repository contains a .agents/ directory, and whenever you are about to write a TODO list, progress log, experiment record, or scratch script.
migrate-to-t9x
Convert an existing project into a t9x workspace: sweep GitHub issues, TODO files, inline TODO/FIXME comments, scattered notes, experiment logs, and loose scripts into .agents/ as tasks, notes, runs, and scripts. Use when a project adopts t9x and its work state lives somewhere else — an issue tracker, markdown files, code comments — or nowhere in particular.
markdown-formatting
Format GitHub Flavored Markdown for human-facing documentation that should render cleanly in both GitHub and Obsidian. Use when creating, editing, rewriting, or reformatting Markdown documents, README files, specifications, research notes, or agent notes where consistent structure, restrained typography, ASCII source, and portable math formatting matter.
manuscript-tasks
Extract embedded work instructions from manuscripts, papers, reports, or source files into t9x tasks, leaving a short @id anchor behind. Use when a human-authored document contains inline requests for writing, derivation, literature search, empirical analysis, or editing — especially explicit `@(word ...)` markers.