Per-process tool-call statistics: a `tool_stats` tool reporting per-tool call counts, error counts, and average durations.
Install
# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)
dsh plugin --profile web add github:disyli/dsh-tool-call-stats
Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time — pnpm blocks those until you allow them, so an install can stop with ERR_PNPM_GIT_DEP_PREPARE_NOT_ALLOWED or ERR_PNPM_IGNORED_BUILDS; dsh prints the exact key to add under allowBuilds in your profile’s pnpm-workspace.yaml, and the install works on the next run. Allowing a build is a trust decision: only install sources you trust, and pin a commit (github:owner/repo#sha).
README
A tiny community plugin for DeepSeek Harness (dsh).
It records every committed tool call in the current harness process and exposes a tool_stats tool, so the model (or you) can ask at any point:
Which tools were used, how often did they fail, and where did the time go?
Not to be confused with the token/cost trackers in the ecosystem (dsh-token-stats, dsh-usage-report, ...) — this plugin counts tool-call dispatches, not tokens. Also unrelated to dsh-tool-stat (a descriptive-statistics math tool).
Status: built and verified end-to-end against the dsh v0.1 developer preview (headless profile, live DeepSeek session): both event listeners record real dispatches and
tool_statsreturns the table below. Upstream interfaces may still change during the preview.
Example output (from the verification run):
tool calls errors avg_ms
---- ----- ------ ------
bash 1 0 23
read 1 0 8
How it works
Everything in dsh is a plugin, and this one is deliberately minimal — three seams, ~100 lines, no build step:
tools/execute(waterfall): wraps each dispatch to record its start time, then delegates vianext().tools/result(observer): counts each committed result per tool name and accumulates duration;result.isErrordrives the error counter.ctx.tools.register(defineTool(...)): exposestool_stats, which returns a canonical JSON array (output.schema) and renders a plain-text table for the model.
All state is in-memory and process-local. Nothing is persisted; restarting dsh resets the counters. When the plugin is unloaded, Cordis disposes both listeners and the tool automatically — no manual cleanup code.
Install
Into a profile (recommended)
dsh plugin --profile <name> add github:disyli/dsh-tool-call-stats
pnpm ≥ 10 will ask you to allow the package before running install scripts (this package has none, but the prompt may still gate the git install). Follow the hint dsh prints, then re-run add.
Verify the layer and start:
dsh --profile <name> --dump-config # shows a "# == dsh-tool-call-stats" layer
dsh --profile <name>
As a local overlay (source checkout)
From a deepseek-harness checkout, point a patch at the local file:
# stats.cordis.yml
- insert:
- id: tool-stats
name: '/absolute/path/to/dsh-tool-call-stats/index.js'
pnpm dsh web --patch ./stats.cordis.yml
Use
Ask the agent things like:
- "Use tool_stats to show tool usage so far."
- "Which tool failed the most in this session?"
Limitations
- Process-local only: counters reset on restart and are not shared across sessions.
- Calls aborted before the registry commits a result are not counted (their timing entries are garbage-collected after 10 s).
- Built against the dsh v0.1 developer preview; upstream interfaces may change ("THERE WILL BE COMPATIBILITY-BREAKING CHANGES").
License
MIT
Links
More in this category
yjh051108/dsh-routing-suite★ 7000
One repository, three parts: a runtime injector for DSH plugin packages (inject, hot-reload, unload, promote a dev staging tool to the front, route self-heal, plus a settings-page plugin manager that lists, unloads and drags folders in to internalize), a task-aware reasoning-mode router agent preset (router-standard / router-spec / router-react), and a graded two-level task protocol whose six tools (commit_star, lock_stage, revise_do, edit_plan, mark_task, redteam_verdict) pin task state to disk. The injector implementation ships in-tree, so the install carries its own behaviour rather than a dependency list.
strukto-ai/mirage#dsh★ 3663
Swaps the filesystem and bash providers for a mirage virtual workspace: file tools and shell commands run over mounted resources (RAM, S3, Redis, Slack, Gmail, Notion, Postgres) instead of the host disk, with per-mount read/write/exec modes, per-command sandbox routing (monty, pyodide, quickjs in process; docker, e2b, daytona remote), and installed CLIs (git, gh, slack, linear, ntn, gws, or one you register) as head words in the virtual terminal.
hust-open-atom-club/oh-dsh★ 325
Community distribution: TUI, desktop, and Web UI as one bundle with layered installation.
weijiafu14/pi2dsh★ 206
Pi Host ABI compatibility engine: after one install, unmodified Pi extensions from npm mount as native DSH plugins with `dsh plugin add <pi-package>`. Verified end to end on stock DSH with pi-mcp-adapter (full MCP manager: OAuth, resources, prompts, MCP Apps, elicitation, sampling), @tintinweb/pi-subagents, pi-code, pi-hermes-memory and pi-background-tasks; `pi2dsh inspect` reports a package's compatibility before installing.
lire1131/dsh-undo-savepoint★ 165
Undo/redo & rollback system for DSH: every config change is auto-snapshotted; undo/redo/restore to any version from the WebUI or the offline CLI/GUI tools (works even when DSH fails to boot).
Fishquito7/dsh-skill-mcp-panel★ 152
Manages DSH skills and MCP servers from the web settings: skill cards with hot enable/disable, workspace scopes, groups, batch migration and drag-and-drop import, plus stdio/HTTP MCP CRUD with connection tests, secret redaction and the unified dsh-panel CLI.
Community comments
Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.