amethystluna/logicprobe
Design document and refactoring plan claim verification: claim enumeration + code base fact check + logic primitive verification (7 structures + 7 adversarial probes), dsh native bundle injection verification discipline gate
Logic Probe (dsh-logicprobe) is a claim-verification skill packaged as a cordis plugin bundle: it checks every verifiable claim in design docs, architecture specs and refactoring plans against the actual codebase, escalating to executable-model verification for behavioural claims. On state-machine models it runs 8 structural checks (S1-S8: reachability, deadlock, liveness, determinism, event completeness, guard completeness, invariant validity, monotonic variables) and 14 adversarial probes (A1-A14: unexpected events, race interleaving, order permutation, pair symmetry, boundary blast, resource injection, minimal counter-example, idempotent replay, leads-to, sequence, atomicity, budget, probability reachability, deadline). It adds data-model verification (DataModelV1), concurrency-risk mining, two-machine composition verification, and export to UPPAAL / TLA+ / PRISM / SPIN inputs. The model is always shown as a transition table and confirmed with the user before running.
Install
dsh plugin --profile web add dsh-logicprobenpm package dsh-logicprobe 0.7.1 (registry re-verified 2026-09-30; repository field back-links to github.com/AmethystLuna/logicprobe, MIT). The README also documents the GitHub source form dsh plugin --profile web add "github:AmethystLuna/logicprobe" and an npx -p @deepseek-ai/dsh dsh plugin ... form when dsh is not installed globally. Restart the profile, then dsh --profile web --dump-config: the id: logicprobe row must appear with enabled: true. The README notes the package name is dsh-logicprobe and the dependency key and dsh.profile.bundles entry must match or the loader fails with ERR_MODULE_NOT_FOUND.
Compatibility
DeepSeek Harness >= 0.1.0-rc.7 (the README re-measured 0.1.5-rc.2 through 0.2.0-rc.2 with per-release evidence in DSH-COMPATIBILITY.md). The Web Plugins-page 'Gate injection' switch requires dsh >= 0.1.7-alpha.1; on older dsh the plugin still loads and injects with the switch simply absent. It also supports Claude Code, Codex CLI, Cursor, Kimi CLI, OpenCode and ZCode via the Agent Skills standard, and Python 3.6+ is optional (only for the automated harness).
Details
- Repo: amethystluna/logicprobe
- Category: Files & Data
- Stars: 1
- Version: npm dsh-logicprobe 0.7.1 (registry re-verified 2026-09-30); MIT
- Last push: 2026-09-19
- First seen: 2026-08-12
Recent updates
The README documents the DSH-native bundle: the skill is discovered by dsh's skill-filesystem provider, and the bundle injects a claim-verification gate into the first model step plus native tools (logicprobe_verify, logicprobe_datamodel_verify, logicprobe_concurrency_scan, logicprobe_compose_verify, logicprobe_export) and a logicprobe:mode context. Configuration keys are enabled (default true), interaction (ask/auto/follow-approval, default follow-approval), and gateContent; the gate switch is editable live from the Web GUI (sidebar Plugins card -> 'Gate injection') without a profile restart.
FAQ
- How do I install Logic Probe in DeepSeek Harness?
- Run dsh plugin --profile web add dsh-logicprobe (or the GitHub source / npx forms), restart the profile, and check dsh --profile web --dump-config shows the id: logicprobe row with enabled: true.
- What does it verify?
- Per the README it enumerates the verifiable claims in a design doc or plan and checks each against the codebase, then runs 8 structural checks (S1-S8) and 14 adversarial probes (A1-A14) on extracted state-machine models, plus data-model, concurrency and composition verification.
- Does it fix code automatically?
- No. The README states it produces structured findings with exact file:line evidence, severity classification, and a correction direction -- it never applies inline fixes.
Alternatives
Jayden-X-L/forkprobe · boomzikazita/dsh-skill-lazy · omdsh-dev/dsh-plugin-skills