EDOS
Ork Meter

All Claude Code plugin versions

RELEASE

Claude Code plugin 0.5.9

Published 2026-11-08 · Release on GitHub

This is the release as published on GitHub, unedited. To install the current plugin, use the install page.

Ork Meter plugin for Claude Code 0.5.9.

Install through the EDOS marketplace (README, "Connect"). It serves this version once the pull request that points it at v0.5.9 is merged (docs/release.md, "Release channel"):

/plugin marketplace add EDOS-Engineering/ork-meter-claude
/plugin install ork-meter@edos

Assets: the plugin tree as installed (zip), each runtime binary, the SBOM, a .sha256 per asset and manifest.json. The binaries are unsigned (docs/release.md, "Signing and Gatekeeper on macOS").

asset SHA256
ork-meter-claude-plugin-0.5.9.spdx.json 48d1cd42c73608a0c4c3d34d864308131676e59d64b47f91e8539117c8397d51
ork-meter-claude-plugin-0.5.9.zip b574450cecfcd0ebcb5d39ab626077807ac76413be67fc8c8181083f5e77cb95
ork-meter-plugin-runtime-darwin-amd64 61d1812c6b372f9d1d4050d038674caded4783cdaf1a72a2e447ab51700bbdb0
ork-meter-plugin-runtime-darwin-arm64 f4003b2cce1c202a28dd774d31b4e4586a1336dcd5faad36aa24bf70eef3bdbe
ork-meter-plugin-runtime-linux-amd64 705a835e3e0fbff83bfa523a400d19c0a823960994a6d0e52c5fecd1ccfcef32
ork-meter-plugin-runtime-linux-arm64 b08bc3a99a54ab76a72902fda6578523cafff84c92791f11aaab7b569af665ae
ork-meter-plugin-runtime-windows-amd64.exe 8961871ca63669169d269e9fa157c53a3e65dc07070fbd6059b8850dc94b8e77

Internal benchmark results

Results for this version only. Run claude-code/2026-11-08-heavy-jail-059: 82 attempts in 41 pairs. Changes are plugin on versus off over valid pairs; a minus sign means the plugin used fewer.

S1 heavy-output (frozen 2026-10-22 task set), effort medium

Non-cached tokens
−27.0%
95% CI
[−40.1, −9.8]
Valid pairs
16 of 16 run
Pass rate OFF
16/16 (100.0%)
Pass rate ON
16/16 (100.0%)
Pass-rate change
0.0 points [0.0, 0.0], McNemar p 1.0
Non-inferior at −5 points
true
Wall time
+5.8% [−4.6, 17.0]
Turns
−1.1% [−13.8, 17.7]
A/A gate
passed

S6 heavy long sessions (Django, SymPy at pinned commits; 14 or 15 scripted turns)

Non-cached tokens
−2.1%
95% CI
[−11.1, 7.8]
Valid pairs
10 of 10 run
Pass rate OFF
9/10 (90.0%)
Pass rate ON
9/10 (90.0%)
Pass-rate change
0.0 points [0.0, 0.0], McNemar p 1.0
Non-inferior at −5 points
true
Wall time
+2.1% [−6.1, 12.2]
Turns
−8.1% [−34.2, 18.2]
A/A gate
flagged: A/A CI [-25.0, -9.7] excludes 0 by more than 3 points

S1 heavy-output: −27.0% non-cached tokens and −21.7% estimated cost, quality unchanged; the saving comes from the 3 of 16 sessions where a lever fired.

S6, the primary suite, is inconclusive: −2.1% [−11.1, +7.8], and its pre-registered A/A noise gate failed, so it is not headlined until that is explained.

Methodology, suites and detailed results