v0.8.0 Released — 100 Tools, Benchmark Analysis, Windows Fixes

[Dave-London]

v0.8.0 is live on npm!

All 14 packages have been published at version 0.8.0.

This release includes everything from v0.7.0 (which was not announced separately) plus additional fixes and analysis work.

---

100 Tools Across 14 Packages (v0.7.0)

Pare has grown from 62 tools and 9 packages to 100 tools across 14 packages.

New packages

Package	Tools	Description
@paretools/github	8	Wraps the gh CLI — pr-view, pr-list, pr-create, issue-view, issue-list, issue-create, run-view, run-list
@paretools/search	3	Wraps ripgrep and fd — search, find, count
@paretools/http	4	Wraps curl — request, get, post, head
@paretools/make	2	Wraps make and just — run, list

Expanded existing packages

Package	New tools
git	+5: tag, stash-list, stash, remote, blame
docker	+4: inspect, network-ls, volume-ls, compose-ps
go	+3: env, list, get
python	+3: pip-list, pip-show, ruff-format
npm	+2: info, search
cargo	+2: update, tree
lint	+2: stylelint, oxlint

Granular tool selection

New PARE_TOOLS and PARE_{SERVER}_TOOLS environment variables let you enable only the tools you need, reducing tool registration noise for agents with limited tool budgets.

---

Comprehensive Benchmark (v0.8.0)

We benchmarked all 100 tools across 148 scenarios, measuring token efficiency against raw CLI equivalents.

Key findings:

• 67% token reduction per session — from 183,289 to 60,254 tokens in a typical 200-call coding session
• 123,035 tokens saved per session at the estimated usage-weighted rate of $4.50/MTok = $0.55/session or $13–27/developer/month
• Top savers: git diff (22K tokens saved per scenario), pip-audit (5.7K), biome-check (4.8K)
• Median latency overhead: -6 ms (negligible — Pare adds no meaningful execution time)

The full Benchmark Summary with methodology, session impact table, cost analysis, and detailed breakdown is available in the repo.

---

Windows Reliability (v0.8.0)

• Fix git log/show format strings: Angle brackets in --format="%an <%ae>" were being interpreted as I/O redirects by cmd.exe when shell: true. Added a shell override to the shared runner so git operations use shell: false on Windows.
• Align CI timeout layers: Vitest testTimeout, inline test timeouts, and runner command timeouts all aligned to 120s to prevent flaky Windows CI failures.

Token Efficiency Improvements (v0.8.0)

• npm: Flatten nested deps with > delimited paths in compact mode, remove resolved URLs from list schema, remove fileCount/unpackedSize from info
• build: Drop redundant errorCount/warningCount from schemas
• lint: Trim diagnostic schema — drop column, fixable, endLine, endColumn
• docker: Cap logs full mode output, truncate container IDs to 12 chars
• git: Merge author+email into single field, deduplicate blame output, add copied field to diff
• test: Restore message field in compact failure output

---

Testing

1,796 tests across 119 files — comprehensive fidelity, integration, security, and compact mode tests for all 14 packages.

Install

npx -y @paretools/git
npx -y @paretools/test
npx -y @paretools/npm
npx -y @paretools/build
npx -y @paretools/lint
npx -y @paretools/python
npx -y @paretools/docker
npx -y @paretools/cargo
npx -y @paretools/go
npx -y @paretools/github
npx -y @paretools/search
npx -y @paretools/http
npx -y @paretools/make