What
/poteto-mode is pstack's hub. Its SKILL.md routes to leaf skills (how, architect, swarm, interrogate, unslop, technical-writing, no-comments, the principle-* leaves) and instructs the agent to read each leaf in full. It gives the leaves by bare name only. poteto-agent.md does the same.
Nearly every pstack skill sets disable-model-invocation: true (all except setup-pstack in 0.15.15). That was introduced in #300 for five skills (how, why, make-bot-ui, typescript-best-practices, unslop) with no stated rationale, and later adopted as the house style for operator-invoked skills: #494's /correct says "the same as the other operator-invoked pstack skills". Commenters on #300 and #308 asked about this, especially for unslop ("Must always apply"), and never got an answer. #379 dropped the flag on unslop alone and was closed as superseded.
How it breaks
- Hidden skills are omitted from the agent's skill list, so the model never receives their paths.
- The Skill tool rejects them unless the user typed that skill's own name this turn. Typing
/poteto-mode does not count as typing /how. The failure is: Skill <name> cannot be used with Skill tool due to disable-model-invocation.
- The same flag blocks preloading the skill into subagents, so a playbook step that spawns a subagent cannot load the leaf either.
- The only fallback is reading the file directly. That works only if the model already knows the install path. In Cursor, the search tools make it worse: Glob and Grep ignore
target_directory values outside the open workspace without error and search the workspace root instead (measured 2026-10-08, Cursor 3.23.23). An agent pointed at a plugin folder silently gets a scan of the repo it is working in, and the lookups fall back to shell commands.
- For plugin-delivered skills, the flag additionally hides them from the
/ palette, contradicting the docs' "only via /skill-name": forum thread, staff-acknowledged 2026-03-24, still unfixed:
https://forum.cursor.com/t/disable-model-invocation-true-completely-hides-plugin-delivered-skills-from-command-palette/155748
Ask
Pick one, or correct me if the intent is something else:
- Make the skills
/poteto-mode routes to model-invocable (or invocable from an already-running poteto session), keeping the flag on side-effect commands.
- If file-read was the intent, add path guidance: poteto-mode should say leaf skills resolve to
../<name>/SKILL.md relative to its own folder and playbooks live under poteto-mode/playbooks/, and poteto-agent should say the same so subagents can resolve them too.
As it stands, the hub cannot discover the skills it is told to route to. Verified against 0.15.15 (upstream main as of 2026-10-08).
Related: #300, #308, #367, #379, #392, #494
What
/poteto-modeis pstack's hub. Its SKILL.md routes to leaf skills (how,architect,swarm,interrogate,unslop,technical-writing,no-comments, theprinciple-*leaves) and instructs the agent to read each leaf in full. It gives the leaves by bare name only.poteto-agent.mddoes the same.Nearly every pstack skill sets
disable-model-invocation: true(all exceptsetup-pstackin 0.15.15). That was introduced in #300 for five skills (how,why,make-bot-ui,typescript-best-practices,unslop) with no stated rationale, and later adopted as the house style for operator-invoked skills: #494's/correctsays "the same as the other operator-invoked pstack skills". Commenters on #300 and #308 asked about this, especially forunslop("Must always apply"), and never got an answer. #379 dropped the flag onunslopalone and was closed as superseded.How it breaks
/poteto-modedoes not count as typing/how. The failure is:Skill <name> cannot be used with Skill tool due to disable-model-invocation.target_directoryvalues outside the open workspace without error and search the workspace root instead (measured 2026-10-08, Cursor 3.23.23). An agent pointed at a plugin folder silently gets a scan of the repo it is working in, and the lookups fall back to shell commands./palette, contradicting the docs' "only via /skill-name": forum thread, staff-acknowledged 2026-03-24, still unfixed:https://forum.cursor.com/t/disable-model-invocation-true-completely-hides-plugin-delivered-skills-from-command-palette/155748
Ask
Pick one, or correct me if the intent is something else:
/poteto-moderoutes to model-invocable (or invocable from an already-running poteto session), keeping the flag on side-effect commands.../<name>/SKILL.mdrelative to its own folder and playbooks live underpoteto-mode/playbooks/, and poteto-agent should say the same so subagents can resolve them too.As it stands, the hub cannot discover the skills it is told to route to. Verified against 0.15.15 (upstream main as of 2026-10-08).
Related: #300, #308, #367, #379, #392, #494