Skip to content

Orchestrating modes

Observation date: October 27, 2026. Playbook counts and port parity are not re-verified every pass — treat specific counts as dated.

An orchestrating mode reads a request, classifies it, and selects a playbook or skill. It depends on a library to do the work. Unlike a composable or lifecycle library, it decides which capability to invoke. See Skill Libraries for those categories.

Lead example: pstack / poteto-mode

pstack is maintained by Cursor engineer "poteto", a React core team member. Its canonical source is the pstack/ directory in Cursor's official cursor/plugins marketplace repository. The README states its priority: "throughput without quality is not a goal I aspire to. If you want to go fast, go deep first."

The poteto-mode skill is installed as /poteto-mode, or /pstack:poteto-mode in a host namespace such as Claude Code's. It reads the constraints in its SKILL.md, classifies the request, and selects a playbook and skill for architecture decisions, parallel work, contested designs, or commit and review checks:

  • architect — cross-boundary design decisions
  • swarm — parallel-agent coverage matrices
  • arena — design bakeoffs between competing approaches
  • interrogate — multi-model adversarial review before shipping

poteto-mode selects other pstack skills based on context; it does not perform their work itself. A lifecycle library's slash commands invoke capabilities directly.

Prior research reported 23 playbooks. That count was not re-verified this pass.

Install paths and port drift

Host Install Caveat
Cursor (canonical) /add-plugin pstack The source of truth — other hosts run a generated or adapted copy
Claude Code /plugin marketplace add cursor/plugins then /plugin install pstack@cursor-plugins Auto-update is off by default for third-party marketplaces and must be enabled manually in the marketplaces tab

Claude Code runs a generated adaptation of pstack from harness/pstack/ at the repository root. CI checks that the generator keeps it in sync with the Cursor source.

Steps requiring Cursor-only features such as Bugbot or Automations are skipped on Claude Code with a one-line note. This is documented behavior, not a failure. Compare unexpected behavior with the canonical cursor/plugins source; synchronized files do not establish feature parity.

The unrelated "potato-mode" name collision

pstack/poteto-mode is unrelated to aryannaio/potato-mode-agent-skill. The latter is a skill for Claude Code and Codex that measures memory, disk, and CPU headroom, then recommends conservative test-runner and build-tool worker limits. It limits resources for the agent's own commands and does not touch unrelated processes.

The similar names refer to different authors, repositories, and functions: poteto-mode routes requests to skills; potato-mode limits resource use.

How this differs from adjacent categories

Compared to... The difference
A composable library (Skill Libraries) You select skills from a composable library. An orchestrating mode selects them based on the request.
A lifecycle library A lifecycle library maps skills to fixed stages you invoke by name (/spec, /build). An orchestrating mode classifies free-form requests and routes dynamically — there's no fixed command per stage.
A native runtime A runtime (Claude Code, Codex CLI, Cursor) executes tool calls and manages sessions. poteto-mode runs inside it and adds library-specific playbook selection.
A methodology (such as Superpowers) Superpowers prescribes an end-to-end sequence, including brainstorming before planning. poteto-mode selects skills per request without owning the full development lifecycle.

When explicit routing helps

Our interpretation follows the three-layer model: as models improve at classifying requests, they may need less explicit routing for simple tasks. Routing may remain useful where playbook selection depends on project-specific judgment, such as contested designs or cross-boundary architecture.

The category is too new to establish how long that advantage will last. Adopt an orchestrating mode for a demonstrated routing need, not as a permanent architectural requirement.