[Bug]: After a fresh auto-restart the new session is told "Post-Compaction Context Recovery" — the agent then reports a compaction that never happened
Nadie ha tomado este issue todavía.
Evaluación
- Dificultad
- 2/5
- Tiempo estimado
- 1-3 horas
- Aptitud para principiantes
- 76/100
- Tipo de issue
- Error
- Claridad
- Bien especificado
- Estado de actividad
- Activo
- Stack tecnológico
- bash, shell
- Área
- developer-experience
Línea de trabajo
Empieza por agents/context/post-compact-resume.sh, especialmente por el manejo de SOURCE_TAG, el contexto emitido, la entrada .budget-status y el comentario de reinicio descrito en el issue. Rastrea cómo recibe el hook el origen y cómo el sentinel de auto-restart identifica una sesión nueva. Actualiza la redacción y el origen registrado para cada caso; el trabajo estará terminado cuando auto-restart se describa como una sesión nueva y compact y resume mantengan descripciones precisas.
Escrito por el modelo de indexación a partir del texto del issue.
Descripción
Bug Description
agents/context/post-compact-resume.sh serves three SessionStart sources (compact, resume, and startup when claude-fw left the .auto-restart-pending sentinel). It reads the source into SOURCE_TAG and uses it to decide whether to run and whether to advance the loop counter, but the text it emits is the same for all three:
- line 143:
CONTEXT="# Post-Compaction Context Recovery (automatic) - line 121:
"source": "post-compact-resume"written to.context/working/.budget-status - line 409: "This context was auto-injected by the session resume hook"
- the block also carries a "Post-Compact Budget Note" that speaks of "this resumed session".
With the fresh restart that is the default since T-3166, the conversation is entirely gone from context: new session, new transcript. That one case is described with the words of the cases where the conversation is summarised (compaction) or restored (resume). The comment at lines 45 to 47 still describes the restart as "claude-fw → claude -c".
Steps to Reproduce
grep -n 'Post-Compaction Context Recovery\|"source": "post-compact-resume"' agents/context/post-compact-resume.shgrep -n 'SOURCE_TAG' agents/context/post-compact-resume.sh— the tag never reaches the title.- Live: start a session with
claude-fw, default restart mode, and let it reach the budget cap. Read the first context block of the new session, then ask the agent what happened.
Steps 1 and 2 were run on 1.8.6 and again on 1.8.11 (e42aa41bf): same lines, same numbers. Step 3 was observed once on 1.8.6.
Expected Behavior
The injected block says what happened: a new session after a budget auto-restart, the previous conversation not in this context, everything below coming from the handover. It gives the identifier of the previous Claude session and the way to reopen it. .budget-status records the real source.
Actual Behavior
The operator asked: "what happened? is the conversation empty?" The agent, reading "Post-Compaction Context Recovery" and a budget status whose claude_session_id was its own, answered that a compaction had taken place in the same session. Both statements were wrong. It found the truth only by reading bin/claude-fw and the process list, and had to retract its first answer.
fw doctor output
AI Agent
Claude Code
Operating System
Linux
Additional Context
Version: 1.8.11 (e42aa41bf), vendored consumer. First observed on 1.8.6; the quoted lines of agents/context/post-compact-resume.sh are unchanged in 1.8.11.
Related: the documents still describe the restart as claude -c (reported separately).
Suggestions:
- Choose the title and the wording from
SOURCE_TAGand the sentinel. - For a fresh auto-restart, give the previous Claude session identifier and mention
claude --resume. - Write the real source in
.budget-status(auto-restart,compactorresume). - Update the comment at lines 45 to 47.
- Lenguaje dominante
- Shell
- Estrellas
- 16
- Forks
- 6
- Merge medio
- 6 h 24 min
- PR fusionados (30 d)
- 5
Preparar el entorno
- Incluye un Dockerfile o un archivo de Docker Compose
- Sin plantilla de pull request
- Leer la guía de contribución
Primeros pasos
- Lee el issue completo y luego la guía de contribución del proyecto.
- Comenta en el issue que vas a ocuparte — evita que dos personas hagan lo mismo.
- Haz un fork del repositorio y trabaja en una rama.
- Abre un pull request que haga referencia al número del issue.
Más de DimitriGeelen/agentic-engineering-framework
-
area:cli bug severity:low status:fixed-on-bleeding-edge triaged
Dificultad 2/5 1-3 horas Aptitud para principiantes 75/100
DimitriGeelen/agentic-engineering-framework#100 · 5 comentarios ·
-
area:cli bug severity:medium status:fixed-on-bleeding-edge triaged
Dificultad 2/5 1-3 horas Aptitud para principiantes 76/100
DimitriGeelen/agentic-engineering-framework#97 · 5 comentarios ·
-
area:docs bug severity:medium status:in-progress triaged
Dificultad 2/5 1-3 horas Aptitud para principiantes 78/100
DimitriGeelen/agentic-engineering-framework#95 · 4 comentarios ·
-
area:watchtower bug severity:low status:released triaged
Dificultad 2/5 1-3 horas Aptitud para principiantes 76/100
DimitriGeelen/agentic-engineering-framework#94 · 4 comentarios ·
-
area:watchtower enhancement severity:low status:released triaged
Dificultad 2/5 Menos de una hora Aptitud para principiantes 75/100
DimitriGeelen/agentic-engineering-framework#91 · 5 comentarios ·
Todos los issues de DimitriGeelen/agentic-engineering-framework
Issues similares
-
Dificultad 2/5 1-3 horas Aptitud para principiantes 78/100
Los mantenedores suelen responder en 1 día
-
Dificultad 1/5 Menos de una hora Aptitud para principiantes 88/100
Los mantenedores suelen responder en 1 día
-
Dificultad 2/5 1-3 horas Aptitud para principiantes 72/100
koalaman/shellcheck#3593 ·
-
Link Checker ReportAbiertoautomated issue report
Dificultad 1/5 Menos de una hora Aptitud para principiantes 78/100
a2aproject/A2A#2381 ·
Los mantenedores suelen responder en 3 días
-
e2e-failure ready-to-code
Dificultad 1/5 Menos de una hora Aptitud para principiantes 82/100
redhat-developer/rhdh-plugin-export-overlays#4485 · 1 comentario ·
Los mantenedores suelen responder en 1 día