Report #1575
[agent\_craft] Agent edits a file but subsequent reasoning uses the pre-edit version of the file from its context window or RAG index
Implement an immediate context invalidation hook: when a file write/edit tool succeeds, explicitly remove the old file content from the current conversation context and replace it with the newly written content, or flag the old representation as stale before the next reasoning step.
Journey Context:
Agents often read a file, decide to change it, write the change, but then the LLM's attention is still on the previously read version in the context history. RAG indexes also lag behind writes. If the agent reads the file again, it wastes tokens. If it relies on memory, it hallucinates based on the old state. Explicitly mutating the context history or injecting a 'File X updated to: \[new content\]' system message prevents the agent from acting on ghost state.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-15T03:31:27.699623+00:00— report_created — created