OpenClaw
OpenClaw proposes a commit-bound inventory for reproducible plugin coverage
OpenClaw's proposed inventory export reads plugin metadata from a specified Git commit rather than a working checkout. A reported validation found 160 manifests and records a digest for reproducible coverage comparisons. The development tool remains open, with ordinary CI and Windows qualification limits explicitly unresolved.
OpenClaw is proposing a more reproducible answer to a basic profiling question: which plugins did a coverage report actually include? PR #155775 adds a JSON inventory export for an explicitly selected Git commit, instead of allowing a sparse, dirty or incomplete working tree to determine the denominator. It is a development and measurement feature, not a change to Gateway behavior.
The proposed collector reads committed Git blobs without executing plugins, excludes documentation-only seeds and records declared plugin surfaces alongside commit and tree identity. A deterministic SHA-256 digest makes the exported inventory comparable across environments. Duplicate identifiers, malformed metadata, symlink metadata and invalid commit selections produce explicit failures rather than partial success. Existing Markdown inventory generation shares the collector and distribution classifier.
The author's current-head evidence reports a dependency-free export validating 160 committed manifests: 63 core, 94 external and three source-only. The report goes beyond counting rows, comparing identities, paths and declared surfaces. That matters when a plugin adds a capability without changing the total plugin count. The same 160-item denominator can otherwise conceal a changed product surface.
The facts
- The development export selects a full commit SHA and requires Git 2.45 or newer. - Plugin metadata is read from committed objects without executing the plugins themselves. - The JSON records commit and tree identities, declared surfaces and a deterministic SHA-256 digest. - The author reports validation of 160 manifests at the inspected head: 63 core, 94 external and three source-only. - Sixteen focused tests reportedly cover real temporary repositories, including dirty working trees and malformed committed metadata. - Ordinary required CI and current-head review remained pending; earlier Windows timeout qualification was not claimed repaired.
Why it matters
A performance comparison needs an auditable denominator. If two runs silently discover different plugins, apparent coverage or startup improvements can reflect missing inputs rather than better behavior. Commit-bound discovery also separates inventory inspection from plugin execution, which avoids making the act of measurement depend on loading every extension into the measuring process.
Current
Checked September 23, 2026. The inspected release baseline is npm latest 2026.9.5; the newer-published GitHub v2026.7.35 is a separate older line. The primary source is open proposal. Tagged release availability and development-source state are separate: a release baseline does not establish that a proposed setting or newly merged feature is in an installed package.
Evidence
Primary: openclaw/openclaw PR #155775 (https://github.com/openclaw/openclaw/pull/155775). Context and implementation: openclaw/openclaw issue #155774 (https://github.com/openclaw/openclaw/issues/155774); openclaw/openclaw — performance.md (source) (https://github.com/openclaw/openclaw/blob/904289cdae0ab23894906762f40fc4084fdd9f43/docs/reference/test/performance.md). The linked source text was inspected; reported experiments and test results remain attributed to their authors.
Operator take
When evaluating the export, retain the selected revision, tree identity and digest with the actual profiling results. Confirm that every expected plugin surface is represented; matching totals alone are insufficient. Keep inventory validation time separate from runtime performance measurements, and do not extrapolate a focused Linux test duration into Windows qualification. The author's evidence makes those distinctions explicitly, which is more useful than a single green badge attached to several unrelated claims.
Caveat
The PR was open at inspection. The 160-manifest export and test results are author-reported evidence tied to the recorded revision, not this run's reproduced measurements. This tool does not establish runtime coverage, fix the earlier Windows timeout or change the Gateway's plugin-loading behavior.
The PR was open at inspection. The 160-manifest export and test results are author-reported evidence tied to the recorded revision, not this run's reproduced measurements. This tool does not establish runtime coverage, fix the earlier Windows timeout or change the Gateway's plugin-loading behavior.