dsh-tool-quota
Run the following command in DeepSeek Harness:
dsh plugin install jwilson411/dsh-tool-quota
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install jwilson411/dsh-tool-quota in DeepSeek Harness to install this plugin; full source is at https://github.com/jwilson411/dsh-tool-quota.
About this plugin
When an agent is handed a search tool, each individual call is perfectly legal. That does nothing to stop it from firing off eighty queries in a single session or pulling back a four-megabyte page and dumping the whole thing into the conversation. dsh-tool-quota closes that gap by enforcing hard caps on both call count and result byte size, scoped per tool and per session.
Two design choices stand out. First, over-quota results are discarded outright rather than truncated: the call fails with a machine-readable code (TOOL_QUOTA_CALLS or TOOL_QUOTA_BYTES), so a caller can branch on the error without parsing prose. Second, the quota is a session-level total, not a rate limiter: there is no time window, no sleep-and-retry, and the allowance does not refill until the session ends or the plugin is reloaded. The plugin also registers a tool_quota_status tool that lets an agent proactively check the remaining allowance for every configured tool, removing the need to discover the cap by hitting it.
It fits developers deploying multi-tool agent workflows who need to prevent any single tool from being over-invoked. The plugin deliberately does not handle spend budgets, token accounting, or permission gating. It does one thing: give every tool a clear per-session usage ceiling and a clear, coded signal when that ceiling is crossed.
Use Cases
- Stop an agent from firing off dozens of search calls or pulling multi-megabyte results that flood the conversation context
- Isolate per-session call allowances so multiple agents sharing a tool do not exhaust each other's quota
- Set a hard byte ceiling on tool results so oversized payloads are discarded rather than silently truncated into the context window
Best For
- Engineers running multi-tool agent workflows who need to prevent any single tool from being over-invoked
- Security or platform teams that require clear usage boundaries and machine-readable error codes when limits are exceeded
- Application developers who want agents to proactively check remaining allowance instead of discovering caps by hitting them
Related Plugins
A unified suite combining hot runtime injection, task-aware thinking-mode routing, and a graded session protocol with red-team gates to sustain model diligence across long-horizon inference.
ModLens is a vision plugin for DeepSeek Harness that gives text-only models sight by reading images pasted directly into chat, with zero-config setup and multiple vision engines.
On-demand vision for text-only DeepSeek Harness agents: built-in free keyless vision chain and 14 vision tools, routing image turns as tool calls to vision models with pixel fidelity, no Python needed, one-command install.
Give text-only models in DeepSeek Harness eyes, enabling image Q&A, long-screenshot OCR, UI restoration, and GUI visual tasks.