p3-stack
AGENTS.md126 starsChanged 38 days agoMITView on GitHub
p3-stack adapts pstack by poteto for T3 Code. Same skills, same language, rewired to T3's native orchestration. A skill reads as if it were written for T3 Code first. Keep pstack's sentences verbatim wherever they still apply. Replace every Cursor mechanism with the T3 Code equivalent below. If you are a delegated worker, your brief names exactly one file. Write only that file, create parent directories as needed, and run no git commands. Report back the file path, its line…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,774 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository51 other files it gives its agents.
- architectSkill
- arenaSkill
- automate-meSkill
- benchmark-checklistSkill
- blast-radiusSkill
- broSkill
- correctSkill
- create-verification-skillSkill
- figure-it-outSkill
- howSkill
- interrogateSkill
- maintain-verification-skillSkill
- make-bot-uiSkill
- no-commentsSkill
- p3-helpSkill
- p3-modeSkill
- principle-attack-the-premiseSkill
- principle-boundary-disciplineSkill
- principle-build-the-leverSkill
- principle-encode-lessons-in-structureSkill
- principle-exhaust-the-design-spaceSkill
- principle-experience-firstSkill
- principle-explain-the-numberSkill
- principle-fix-root-causesSkill
- principle-foundational-thinkingSkill
- principle-guard-the-context-windowSkill
- principle-laziness-protocolSkill
- principle-make-operations-idempotentSkill
- principle-migrate-callers-then-delete-legacy-apisSkill
- principle-minimize-reader-loadSkill
- principle-model-the-domainSkill
- principle-never-block-on-the-humanSkill
- principle-outcome-oriented-executionSkill
- principle-prove-it-worksSkill
- principle-redesign-from-first-principlesSkill
- principle-separate-before-serializing-shared-stateSkill
- principle-sequence-verifiable-unitsSkill
- principle-subtract-before-you-addSkill
- principle-test-behavior-not-implementationSkill
- principle-type-system-disciplineSkill
- recallSkill
- reflectSkill
- setup-p3Skill
- show-me-your-workSkill
- swarmSkill
- tddSkill
- teachSkill
- technical-writingSkill
- typescript-best-practicesSkill
- unslopSkill
- whySkill
The file
We split this AGENTS.md into the rules it gives an agent
5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 9
If you are a delegated worker, your brief names exactly one file.
- Distinctive: line 37
Keep these names exactly.
- Very specific: line 44
Use T3 tool names exactly: delegate_task, task_status, watch_pull_request, link_pull_request, schedule_task, t3_thread_launch, …
- Distinctive: line 46
Never bullet a Cursor tool into a skill, even as a fallback.
- Very specific: line 68
Every step names a T3 mechanism or a vendor-neutral habit.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
p3-stack contributor guide
p3-stack adapts pstack by poteto for T3 Code. Same skills, same language, rewired to T3's native orchestration.
A skill reads as if it were written for T3 Code first. Keep pstack's sentences verbatim wherever they still apply. Replace every Cursor mechanism with the T3 Code equivalent below.
Delegated workers
If you are a delegated worker, your brief names exactly one file. Write only that file, create parent directories as needed, and run no git commands. Report back the file path, its line count, and the changes you made to pstack's version.
Cursor to T3 Code map
| pstack | p3-stack |
Task tool, subagent_type, run_in_background | delegate_task with a self-contained brief; mode: "async" for background work; task_status to drain; task_cancel to stop |
Model slugs (claude-opus-5-5-max, grok-4.7-xhigh-fast, gpt-5.6-sol-max) | Named roles resolved from orchestrator_capabilities by setup-p3 (role names below). Never hardcode a slug. |
"cloud agent", environment: "cloud", cloud_base_branch | A delegated child task. Use t3_thread_launch only when the worker needs its own top-level thread, worktree, or branch. |
| "each worker gets its own worktree or branch" | t3_thread_launch with workspaceStrategy: {"type":"worktree","baseRef":...,"branch":...}, {"type":"existing_worktree","worktreePath":...}, or {"type":"root"}. One writer per worktree. |
/loop, poll loops, scripts/watch-pr | watch_pull_request: T3 watches the PR and wakes the thread on checks finishing, new comments, or conflicts. Arm it, end the turn, triage on wake. For other cadences use schedule_task. |
| Hourly audit ticks, recurring runs | schedule_task with {"type":"interval","everyMs":...} or {"type":"fixed_time","timeOfDay":...,"weekdays":[...]}; run_scheduled_task_now for an immediate run. |
Stack bookkeeping with gt, manual PR lists | link_pull_request for every layer, immediately; list_thread_pull_requests returns the stack bottom to top. |
agent-transcripts/*.jsonl mining | t3_thread_read, t3_thread_search; t3_queue_list/t3_queue_read/t3_queue_edit/t3_queue_reorder/t3_queue_cancel/t3_queue_promote_to_steer for queued input. |
AskQuestion | Ask plainly in the thread. t3_pending_request_list/t3_pending_request_read/t3_pending_request_respond answer questions in other threads. |
control-ui, control-cli (from cursor-team-kit) | preview_* browser tools and device_* simulator tools; preview_recording_start/preview_recording_stop and device_screenshot for evidence; browser.preview shows a file to the user. |
~/.cursor/rules/pstack-models.mdc, /add-plugin | setup-p3 writes p3-models.md; install is a symlink of skills/* into ~/.agents/skills/ (see README). |
poteto-mode, poteto-agent, poteto-help, setup-pstack | p3-mode, agents/p3-agent.md, p3-help, setup-p3. |
| Cursor custom modes ("press option+enter") | Skills are invoked by name (/p3-mode). No mode toggles. |
Everything else in pstack transfers as written: the principles, the playbook shapes, the brief template, the reply style, the autonomy rules. Adopt it.
Role names
Code work reads its line in p3-models.md: feature, refactoring, bug-fix, perf-issue, hillclimb. Prose and judgment read judgment and prose; the hardest changes read hardest tasks.
Panels and investigators: how explorer, how explainer, why investigators, why synthesizer, arena runners, arena cross-judge pool, swarm workers, architect runners, interrogate reviewers, reflect tooling, reflect judgment, divergent, synthesizer.
Keep these names exactly.
Writing rules
- Thin over complete. Keep the shortest file that still changes behavior. Cut Cursor-only edge-case protocol and forge-specific branches before cutting any rule.
- pstack's voice: terse, declarative, lowercase-friendly, no filler.
- Frontmatter:
name, one-linedescription,disable-model-invocation: truewhere the source has it. Drop Cursor UI fields (icon,color,mode,reminder). - Use T3 tool names exactly:
delegate_task,task_status,watch_pull_request,link_pull_request,schedule_task,t3_thread_launch,t3_thread_read,t3_thread_search,preview_*,device_*. - Delegation briefs are self-contained. T3 child agents get only the brief, never the parent's context. Say this where pstack says "every brief stands alone".
- Never bullet a Cursor tool into a skill, even as a fallback.
Layout
skills/<name>/SKILL.mdis a skill.skills/principle-*/SKILL.mdare the short principles.skills/p3-mode/SKILL.mdis the mode router; its playbooks live inskills/p3-mode/playbooks/.agents/holds prompt briefs for delegated roles (p3-agent.md,comment-sicko.md), not harness agent definitions.
T3 tool quick reference
- Delegation:
delegate_task,task_status,task_cancel,orchestrator_capabilities - Threads:
t3_thread_launch,create_threads,t3_thread_list,t3_thread_read,t3_thread_search,t3_thread_send,t3_thread_wait,t3_thread_fork,t3_thread_merge_back,t3_thread_interrupt,t3_thread_organize,t3_thread_update,t3_thread_configure - Worktrees:
t3_worktree_status,t3_worktree_list,t3_worktree_handoff - Pull requests:
link_pull_request,list_thread_pull_requests,watch_pull_request,unwatch_pull_request,unlink_pull_request - Schedules:
schedule_task,list_scheduled_tasks,update_scheduled_task,delete_scheduled_task,run_scheduled_task_now - Queue:
t3_queue_list,t3_queue_read,t3_queue_edit,t3_queue_reorder,t3_queue_cancel,t3_queue_promote_to_steer - Preview:
preview_open,preview_navigate,preview_snapshot,preview_click,preview_type,preview_press,preview_scroll,preview_wait_for,preview_status,preview_resize,preview_set_appearance,preview_recording_start,preview_recording_stop,t3_preview_list,t3_preview_close - Devices:
device_list,device_open,device_screenshot,device_close - Pending requests:
t3_pending_request_list,t3_pending_request_read,t3_pending_request_respond
Definition of done for a skill
- Every step names a T3 mechanism or a vendor-neutral habit. Grep for
Cursor,Task tool,subagent_type,cloud agent,.cursor,agent-transcripts,AskQuestion,icon:,claude-,grok-,gpt-,/loop. None may remain. - It is shorter than the pstack source unless the T3 rewrite genuinely needs the lines.
- Frontmatter matches the rules above, and references to other skills resolve to files in this repo.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

