Pstack

pstack

Lauren Tan's poteto-mode, recut for this catalog. Match the task to a playbook. Copy the playbook steps into the todo list verbatim. Route to the other skills as the steps fire. Name each principle that shaped a decision.

This is an execution router for the task the user selected. Invoke callable companion engines within that scope. Pass the requested target, authorized mutations, report-only restrictions, and host/provider limits to each delegate. Loading a playbook never grants additional authority. Recommend an explicit-only entry point when its separate workflow is needed.

Contract

Inputs:

• A non-trivial task: bug, feature, investigation, refactor, perf, shipping, or a long unattended run

Outputs:

• A matched playbook, a todo list of its steps, and the playbook's named reply
• Principle citations that trace to a real choice

Creates/Modifies:

• Whatever the matched playbook authorizes

External Side Effects:

• Network, git, and GitHub writes only when the playbook's landing or review steps require them

Confirmation Required:

• Irreversible writes: force-push to shared branches, deploys, data deletion, customer messages

Delegates To:

• how, why, architect, arena, swarm, interrogate, tdd, figure-it-out, show-me-your-work, blast-radius, teach, recall
• deslop, no-comments, technical-writing, review-dispatch, worktree, and skill-creator when their steps are authorized
• Recommend finishing-a-development-branch or git-cleanup for their separate explicit workflows

Installation and Companion Resolution

This is the canonical Shipshit integration of Open Pstack and original Pstack. Resolve companions through this distribution's active catalog. Installed resources require neither upstream plugin. Report missing or ambiguous companions before relying on them. Resolve resources relative to each selected installed skill.

Read provider dispatch before launching a configured role. Read optional adapters when configuring a harness; preserve its canonical role sheet and existing generated adapters. The source ledger records exact upstream pins, adaptations and platform-specific capabilities. Optional adapters remain dormant until explicitly configured.

For prose, load references/prose-slop.md from the selected deslop skill.

Start

Open a todo list whose first item is reading references/principles.md in full. Then copy the matched playbook's steps in verbatim, before any task-specific todos. A step you skip stays listed with skip: <reason>.

Non-negotiables

• Nontrivial change, architecture decision, or "are we sure?" → the how skill.
• About to ask "which approach" or "how should I" → classify first. If the answer is a fact you could observe, sketch it via the Prototype playbook instead of asking. Reserve the question for a genuine product or preference call.
• Any code → name the data shape first (Model the Domain).
• Code crossing a function boundary → the architect skill.
• Parallel fan-out → swarm for coverage, races, and partitions. arena for bakeoffs with a base and grafts.
• Contested design → the interrogate skill before shipping.
• Docs, RFCs, READMEs, PR descriptions, or commit messages → technical-writing, then the prose-slop catalog.
• Before review → name no-comments.
• PR-status request → the Babysit playbook. Never triggered by merely opening a PR.
• Asked to land a green stack → the Shipping playbook. Green is not safe.
• Review-bot comments → triage per references/bugbot-triage.md.
• Long, autonomous, or multi-phase work → a decision trail via show-me-your-work.
• Large or cross-cutting work, or work the user reviews after stepping away → figure-it-out even when a narrower playbook fits.

Capability tiers

Orchestrators speak in tiers. The consuming repo's routing block maps tiers to models. Never name a concrete model.

• Fast cheap tier — fan-out finders, coverage workers, mechanical edits.
• Strong instruction-following tier — specified sequences to execute to the letter.
• Strongest judgment tier — synthesis, lead verdicts, contested design, cross-cutting architecture.

Spawn workers on the fast cheap tier unless the step needs judgment or letter-perfect execution. Review every delegate's diff yourself. Do not pass through a self-report. A second opinion is the same prompt on a different tier or family.

Give each writer its own worktree or branch. File pointers, not inlined dumps.

Autonomy

Reversible work proceeds. Pause for irreversible writes. "Don't stop" / "going to bed" / "run until done" keeps going. No is an acceptable answer. Decline scope that does not earn its place.

Writing the reply

Write the reply clean as you draft it.

• Short declarative sentences. One thought per sentence.
• Prefer periods over dashes and mid-sentence colons.
• Frame impact for the consumer and the maintainer before implementation detail.
• Never fabricate a link, citation, or transcript reference.
• Apply references/prose-slop.md from the selected deslop skill directory.

Keep a comment only for a non-obvious why the code cannot show.

Playbooks

Match the task, open the file, copy its steps.

• Investigation. Read-only how/why. playbooks/investigation.md
• Bug fix. Reproduce, root-cause, fix with runtime evidence. playbooks/bug-fix.md
• Perf issue. One measured slowness. playbooks/perf-issue.md
• Hillclimb. Sustained metric loop. playbooks/hillclimb.md
• Runtime forensics. Live-instrumentation diagnosis. playbooks/runtime-forensics.md
• Trace forensics. Captured profile diagnosis. playbooks/trace-forensics.md
• Feature. New behavior from a named data shape. playbooks/feature.md
• Refactoring. Behavior-preserving reshape. playbooks/refactoring.md
• Prototype. Throwaway sketch to decide. playbooks/prototype.md
• Visual parity. Pixel-exact match. playbooks/visual-parity.md
• Authoring a skill. playbooks/authoring-a-skill.md
• Eval. Blind candidate comparison. playbooks/eval.md
• Babysit. Drive a PR to merge-ready. playbooks/babysit.md
• Shipping. Independent verify, then land. playbooks/shipping.md
• Autonomous run. Drive to a predicate. playbooks/autonomous-run.md
• Orchestrate. Standing multi-PR program. playbooks/orchestrate.md
• Autopilot-full. Independent PRs to merged. playbooks/autopilot-full.md
• Autopilot-stack. Linear stack the operator lands. playbooks/autopilot-stack.md
• Session pickup. Resume prior work. playbooks/session-pickup.md
• Pause safely. Clean stop. playbooks/pause-safely.md
• Multi-phase plan. Checklist, no implementation. playbooks/multi-phase-plan.md
• Worktree cleanup. Safety-gated prune. playbooks/worktree-cleanup.md
• Opening a PR. End of every other playbook. playbooks/opening-a-pr.md

Automate Me procedure

Read automate-me procedure when creating or updating a personal mode from the user’s working conventions. Apply the authorized scope and mode of this entry point to every step. Resolve other skills through this distribution’s active catalog; resolve resources relative to the installed skill directory.

Babysit procedure

Read babysit procedure when selecting a PR monitoring workflow; the canonical Babysit playbook owns monitoring modes. Apply the authorized scope and mode of this entry point to every step. Resolve other skills through this distribution’s active catalog; resolve resources relative to the installed skill directory.

Bro procedure

Read bro procedure when the user asks to restate the previous answer more simply. Apply the authorized scope and mode of this entry point to every step. Resolve other skills through this distribution’s active catalog; resolve resources relative to the installed skill directory.

Poteto Mode procedure

Read poteto-mode procedure when selecting and running an engineering playbook. Apply the authorized scope and mode of this entry point to every step. Resolve other skills through this distribution’s active catalog; resolve resources relative to the installed skill directory.