Hacktoberfest 2026: los issues que los mantenedores marcaron para octubre, abiertos y aptos para principiantes. Explorar issues de Hacktoberfest

pstack: in Claude Code, poteto-mode cites principles without reading their SKILL.md files

Abierto Apto para principiantes
#726 1 comentario 0 reacciones 0 asignados Ver en GitHub

Los mantenedores suelen responder en 1 día

Nadie ha tomado este issue todavía.

Evaluación

Dificultad
2/5
Tiempo estimado
1-3 horas
Aptitud para principiantes
65/100
Tipo de issue
Error
Claridad
Bien especificado
Estado de actividad
Activo
Stack tecnológico
typescript

Línea de trabajo

Empieza por las instrucciones de poteto-mode SKILL.md descritas en el issue, especialmente las secciones Non-negotiables y Principles, y compáralas con la redacción propuesta. Actualiza las instrucciones para que exijan explícitamente leer el adaptador y el leaf skill pertinente antes de la primera respuesta. Comprueba que la lectura del adaptador siga estando limitada a Claude Code y que la lectura del leaf skill se aplique en Cursor; el trabajo estará terminado cuando las instrucciones exijan estos pasos y conserven la regla de citación.

Escrito por el modelo de indexación a partir del texto del issue.

Descripción

Problem

In Claude Code, /pstack:poteto-mode runs almost never read the files that the skill tells the agent to read. SKILL.md line 11 says to read docs/harness/claude-code.md once per session. Line 17 says to cite only principles whose leaf SKILL.md the agent read this session. The agent skips both reads. It then does one of two things:

  • It names principles from the inline index without reading them. That breaks the "cite only what you read" rule.
  • It names no principles at all, so the Principles section has no visible effect.

From the user's side, poteto-mode follows its principles on some days and ignores them on others.

Evidence

I audited 18 poteto-mode runs across 4 Claude Code sessions (pstack 0.15.25 and 0.15.27, Claude Code 2.1.304, Opus 5.5):

  • Every run injected the full skill body, so the skill loaded correctly.
  • The agent read docs/harness/claude-code.md in 1 run.
  • The agent read a principle-*/SKILL.md file in 0 runs and a playbook file in 0 runs.
  • One session named principles in its replies anyway. The other three named none.
  • One run spawned general-purpose where the adapter maps delegates to pstack:poteto-agent.

poteto-mode, poteto-agent, and the principle-* skills are identical in 0.15.25 and 0.15.27, so the version change did not cause this.

All 24 principle-* skills set disable-model-invocation: true, so the Skill tool refuses them. The only way to load a principle file is to Read it by path. The current wording states the citation rule but never makes the read a step, so the agent treats the inline index as enough.

Proposed fix

Make the reads an explicit first step in the Non-negotiables:

-The Principles section below grounds every trigger. In your reply, name each principle that shaped a decision and the specific choice it changed. Cite only principles whose leaf SKILL.md you read this session.
+The Principles section below grounds every trigger. Before your first reply in a run, `Read` the adapter above if you have not this session, then `Read` `../principle-<name>/SKILL.md` for each principle you expect to shape a decision. The index below is a pointer, not a substitute. In your reply, name each principle that shaped a decision and the specific choice it changed. Cite only principles whose leaf SKILL.md you read this session.

The adapter clause applies only in Claude Code, because line 11 scopes the adapter to Claude Code. In Cursor the change adds only the leaf read, which line 17 already requires.

Test

I applied the patch locally to 0.15.27 and ran:

claude -p "/pstack:poteto-mode avg() in stats.py crashes on an empty list. Propose a fix, don't edit files."

The agent read docs/harness/claude-code.md, then principle-fix-root-causes/SKILL.md and principle-boundary-discipline/SKILL.md. Its reply named both principles and said what each one changed. That is one run, not a benchmark.

Related

  • #629 adds Claude Code support but does not change line 17.
  • I haven't tested whether poteto-agent delegates have the same gap. Their fallback search for SKILL.md looks only in ~/.cursor/plugins/.
Lenguaje dominante
TypeScript
Estrellas
12.5k
Forks
1.2k
Merge medio
10 h 48 min
PR fusionados (30 d)
128

Preparar el entorno

Este proyecto no incluye contenedor de desarrollo, Dockerfile ni guía de contribución, así que la configuración corre por tu cuenta: empieza por su README y consulta nuestra guía para la primera contribución para los pasos generales.

Primeros pasos

  1. Lee el issue completo y luego la guía de contribución del proyecto.
  2. Comenta en el issue que vas a ocuparte — evita que dos personas hagan lo mismo.
  3. Haz un fork del repositorio y trabaja en una rama.
  4. Abre un pull request que haga referencia al número del issue.

Más de cursor/plugins

Todos los issues de cursor/plugins

Issues similares

Más issues de TypeScript

Recibe los nuevos issues en tu correo

Un resumen breve de issues de GitHub para principiantes.