VeriClaw 爪印
Apple-native correction companion for OpenClaw
AI agent correction, AI纠偏, and role-drift diagnosis for OpenClaw on Apple-native surfaces.
Turn AI hallucination, AI假完成, overreach, and broken professional-role contracts into evidence, diagnosis, intervention, agent evaluation, verification, and casebook learning.
If you are already inside OpenClaw, start from the ClawHub skill or plugin route so the action becomes a real install, not just another anonymous ZIP pull.
If the official ClawHub skill genuinely helps on a real workflow, go back to that same official skill page and leave the star or short comment there so the public signal stays concentrated on the canonical VeriClaw listing.
clawhub install vericlaw
openclaw skills install vericlaw
GitHub release for the standalone app
The cleanest growth actions are the ones that match intent: counted OpenClaw installs on the ClawHub route, direct app downloads on the release route, and GitHub stars for people who want to follow the project before installing.
If you are already using OpenClaw, open the official skill page or
run clawhub install vericlaw so the install happens
on the canonical VeriClaw surface. If it proves useful, leave the
star or short comment on that same skill page instead of on a
mirror or summary page.
If you want the OpenClaw-facing companion shell and bundled discovery skill, use the plugin route instead of treating the app ZIP like an install step.
Use the download page only for the standalone app, and use GitHub when the goal is to watch releases, browse source, or star the project before you install.
VeriClaw is not another monitoring-only surface. It is a correction workspace for teams that need to decide what happened, what should happen next, and whether the fix actually held.
Capture what the assistant actually did before context disappears into another incident thread or dashboard screenshot.
Frame failures as broken professional-role contracts when that helps teams move from confusion to a concrete intervention faster.
Close the loop only after the prescribed correction is tested and the reusable lesson is written back into the casebook.
This repository is the public launch pack for the VeriClaw story: screenshots, positioning, release copy, license and attribution materials, and the GitHub-first launch surface.
VeriClaw now has a cleaner split between the public app download, the OpenClaw companion plugin, the bundled discovery skill, and the source repository. If the goal is a real OpenClaw install, start from the skill or plugin route instead of the app ZIP.
Use the public download page for the real app ZIP, checksums, and release path.
Use the
ClawHub plugin page
when you want the OpenClaw-facing companion entry instead of the
standalone app bundle. The current published companion package on
ClawHub includes the bundled vericlaw discovery
skill, so this is the right route when you want the companion
surface to count as an actual OpenClaw install.
Installing the plugin also exposes the `vericlaw` companion skill inside OpenClaw's Skills surface for canonical routing and product discovery.
Use the
ClawHub skill page
or run clawhub install vericlaw for the standalone
discovery skill route. If you are already inside OpenClaw, use
openclaw skills install vericlaw. The
skill guide explains skill vs plugin
vs app download, and this route is usually the shortest path to a
counted skill install.
Native OpenClaw install commands are
openclaw skills install vericlaw for the skill and
openclaw plugins install clawhub:vericlaw for the
companion plugin route.
If the search phrase is `AI纠偏`, `幻觉纠偏`, `假完成诊断`, or
`未做却说做了`, start from the matching Chinese route page and
keep the install CTA at clawhub install vericlaw or
openclaw skills install vericlaw.
Point people to the GitHub repository, landing page, or download page so the `VeriClaw` name stays attached to the official source.
The page and repository copy intentionally speak to the problems people search for: AI agent supervision, LLM correction, hallucination remediation, role-drift diagnosis, and Apple-native oversight around OpenClaw.
For operators who need more than logs when a production assistant drifts or overreaches.
A correction-first story for teams evaluating supervision workflows instead of monitoring-only observability.
Evidence, intervention guidance, and verification for moments when a bot invents, guesses, or closes work falsely.
A clearer route for teams that describe the problem as output checking, evidence review, or verification before sign-off.
Another useful route for review sites and tooling directories that classify the problem as QA rather than correction.
VeriClaw strengthens OpenClaw as a native companion surface rather than pretending to replace the runtime and gateway.
These linked surfaces are the cleanest machine-readable and citation-ready entry points when a plugin directory, review page, recommendation site, or another skill wants to understand what VeriClaw is.
Canonical plugin and bundled skill explanation for `OpenClaw plugin`, `ClawHub plugin`, and `VeriClaw skill` queries.
Category page for `AI agent correction`, `LLM supervision`, and `hallucination remediation` recommendation intent.
Category page for `AI hallucination`, `AI幻觉纠偏`, `AI纠偏`, and correction-first verification after drift.
Chinese umbrella page for `AI纠偏`, `AI纠错`, `AI校正`, `模型纠偏`, and nearby correction intent.
Chinese exact-match page for `幻觉纠偏`, `幻觉修复`, and `工作幻觉` search intent.
Category page for `AI假完成`, `AI以为自己工作了`, `agent evaluation`, and `verification workflow` intent.
Chinese exact-match page for `假完成诊断`, `AI假完成`, and `AI以为自己工作了` intent.
Natural-language failure page for `未做却说做了`, `明明没做却说做了`, and `装作做完了`.
Copy-ready blurbs, install lines, and review snippets for directories, newsletters, skill roundups, and recommendation sites.
Official listing fields, categories, tags, and safe claims for plugin directories, tool catalogs, and recommendation sites.
Official route index for AI recommendation systems, directories, reviewers, and other machines choosing the best VeriClaw entry.
Long-tail route for `OpenClaw companion`, `OpenClaw correction`, and OpenClaw-native install discovery.
Use-case route for incident response, post-failure review, and correction-first recovery around AI agent failures.
Focused route for `agent supervision`, `AI supervision`, and `AI监督` discovery intent.
Focused route for `agent audit`, `AI output review`, `LLM QA`, and evidence-first operator review intent.
Focused route for `agent review`, `AI work verification`, and claim-checking language used in broader tool roundups.
Long-tail route for `agent quality control`, `correction-first supervision`, and verification-first review.
Focused route for `LLM QA`, `AI output QA`, and evidence-based quality review terminology.
Focused route for `AI output verification`, `verify AI output`, `result checking`, and evidence review intent.
Use-case route for workflow repair, recovery loops, and reopening AI work that failed or closed too early.
Natural-language route for `bot correction`, `bot verification`, and `bot drift recovery` searches.
Use-case route for escalation, human review, and correction-first intervention before trusting AI completion.
Category page for `agent evaluation`, `drift verification`, and evidence-first evaluation after correction.
Category page for `verification workflow`, `correction loop`, and `verification before completion` intent.
Comparison page for people evaluating generic monitoring tools against correction-first workflows.
Comparison page for `agent observability`, `AI observability`, and correction-first recovery evaluation.
Comparison page for teams deciding between evals-style quality measurement and operational correction loops.
Comparison page for `hallucination detection` versus correction-first remediation and verification.
Official facts, screenshots, naming guidance, and canonical links for directories, reviewers, and recommendation sites.
Official clarification for `VeriClaw MCP`, `is VeriClaw an MCP server`, and other directory pages that need truthful labeling.
Broad-intent route for `hallucination checker`, `AI hallucination review`, and correction-first recovery wording.
The public surface is GitHub-first today, but real usage grows fastest when OpenClaw users install from the canonical ClawHub route first and only use the release ZIP when they truly want the standalone app.
The repository already includes shareable launch assets and source-use guidance, even while Apple-native distribution hardening continues in parallel.
