AI Agent Hub
Back to plugins
⚙️

dsh-fork-inbox-guard

Workflow Updated 2026.09.16

Run the following command in DeepSeek Harness:

dsh plugin install azazo1/dsh-fork-inbox-guard

Paste the following prompt into your AI chat to install this plugin:

Run "dsh plugin install azazo1/dsh-fork-inbox-guard" in your DeepSeek Harness terminal to install. Source: https://github.com/azazo1/dsh-fork-inbox-guard

About this plugin

When forking a session in DeepSeek Harness, a subtle but recurring problem trips up heavy workflow users: the child session silently inherits any pending, unconsumed prompt still sitting in the source session inbox. The result is that the child executes that stale message first, your freshly typed prompt gets pushed to the next turn, and the same command may run twice.

dsh-fork-inbox-guard hooks into the agent/session-start event and performs a precise three-condition check: whether the startup is a fork path, whether the session header carries a seed flag, and whether the inbox actually holds pending messages. Only when all three hold does it call inbox.clear() and write an audit record marked with outcome canceled, leaving an explicit inherited-then-voided trail in the child session log. When the conditions are not met it is a true no-op, leaving resume flows and brand-new sessions completely untouched. All entry points are covered: sidebar Fork, message-menu Branch, subagent fork provider, agent-team fork context, and raw sessions.fork() calls.

If you rely on Harness for multi-agent collaboration, branch-based experiments, or high-frequency fork-and-debug cycles, this plugin eliminates the ghost-prompt problem of duplicate execution and response delay. It operates only on the Host side, registers no Client module, and requires no frontend rebuild. Install and restart to activate.

Use Cases

  • Fork a session and make the child respond to your new prompt immediately instead of replaying a stale source message
  • Prevent the same command from re-executing in the child due to inherited inbox state
  • Keep each turn input clean and auditable during multi-agent branch debugging

Best For

  • Harness developers who fork sessions frequently for experimentation and debugging
  • Teams relying on multi-agent collaboration and branched workflows
  • Power users who need precise prompt lifecycle tracking across session boundaries