Cross-session, cross-CLI memory for AI coding assistants. Mines transcripts from 10 supported CLI clients. Surfaces operator identity / decisions / bans / corrections / goals / voice via 26 MCP tools, 6 hooks (SessionStart signpost + SessionStart compact-restore + UserPromptSubmit retrieval + Stop/PostCompact re-index + PreCompact continuity-seed), 15 slash commands, and 3 skills. Hybrid FTS5 + ollama dense recall (qwen3-embedding:0.6b, default-on; TOTAL_RECALL_EMBED_MODEL unset is correct). Product ollama auto-provisions refine (qwen3.5:2b; TOTAL_RECALL_LLM_PROVIDER=none to disable). Worktree-aware project scoping; post-compaction coding-continuity packet. The operator becomes the source of truth; the model stops re-asking what they already told it.
3 skills · 15 commands · 1 MCP server · hooks
Falsification-first investigation workflow: convert every assertion into a labeled falsifiable hypothesis, predict before measuring, run controlled experiments, verify findings adversarially (REFUTE-first), and persist verdicts in a hypothesis ledger so killed ideas are never re-attacked
1 skill · 7 commands · 5 agents · hooks
Seamless context continuity across Claude Code and Grok compaction. Multi-layer background capture (continuous tool-call capture, mechanical PostCompact handoff, async Opus 4.7 enrichment at --effort max, async Stop-hook refinement, and preemptive snapshot before the next compact) keeps the agent working state intact across every compaction and resume. All summarization is isolated from CLAUDE.md/auto-memory and invisible to the user.
1 skill · 8 commands · 1 agent · 1 MCP server · hooks
Ultra-compressed communication mode. Cuts ~75% of tokens while keeping full technical accuracy by speaking like a caveman.
7 skills · 3 agents · hooks
Detects when Claude drifts away from your active output contract (a terse persona, hard formatting/length rules, an in-character voice) and quietly steers it back. Each assistant turn is scored by a deterministic, dependency-free engine; a status-line badge shows live drift and the next prompt gets a one-shot correction nudge when the previous reply broke contract.
1 skill · 6 commands · 1 MCP server · hooks
A ground-first reflex: the agent judges (from its own training, any domain) when a request is complex/irreversible, grounds before acting, and is hard-blocked from the destructive step until it has.
1 skill · 6 commands · 1 agent · hooks
Simplify, then add lightness. Removes AI-generated code fat while preserving functionality, then benchmarks the speedup. Multi-pass: measure weight -> trim -> verify equivalence -> benchmark.
3 skills · 2 agents · hooks
Makes AI work like AI — no human time estimates, no asking for accessible data, no emotional labor.
1 skill · 4 commands · hooks
Enforce full agent authority on reversible work: run commands yourself, never yield instructions back to the operator. Stop hook scores yield-back turns; next prompt gets a one-shot drive correction.
1 skill · 3 commands · hooks
Installs a standing assumption-testing prior: before adopting any borrowed component/runtime/algorithm or accepting any asserted limit ('compute-bound', 'you need X', 'that's how everyone does it'), the model names it as a falsifiable claim and checks whether the constraint is actually load-bearing in THIS regime or cargo-culted from another. Fires every turn via a UserPromptSubmit hook, primed at SessionStart. The external sibling of flip-the-script: that distrusts your own recall, this distrusts inherited assumptions. Routes a broken dogma to invention; a surviving one is a logged result. Calibrated, not contrarian — reuse leaves that aren't the moat.
1 skill · 1 command · hooks
Routes any decision, problem, or approach through Dave Snowden's Cynefin framework before you act: read the constraints, place the domain (Clear / Complicated / Complex / Chaotic — or admit Disorder), and apply the matching response (sense-categorize / sense-analyze / probe-sense / act-sense). Fires a conditional UserPromptSubmit nudge on decision/problem/incident language, primed at SessionStart. Enforces sense-making not categorization (never a 2x2), a safe-to-fail probe portfolio in the Complex domain, and the Clear-to-Chaotic cliff warning.
1 skill · 1 command · hooks
Installs a standing self-distrust prior: the model treats its own unverified recall as stale and overconfident by default, and routes any version/API/current-state/exact-number/long-tail claim to EXTERNAL ground truth (docs, search, read-the-source, measure, or a fresh-context check) before asserting. Fires deterministically every turn via a UserPromptSubmit hook, primed once at SessionStart. Confidence is not evidence; internal second-guessing alone does not fix a wrong recall — the fix is external verification. Calibrated, not paralytic: distrusts hard where recall is unreliable, stays quiet on stable well-known facts.
1 skill · 1 command · hooks
Fires the wait-free/lock-free doctrine exactly when you touch concurrency code: a PreToolUse hook inspects Write/Edit content and, when it sees a lock, atomic, CAS, fetch-add, spawn, channel, or CUDA sync, injects (non-blocking) the Herlihy checklist — pick the primitive by consensus number, fix the linearization point, claim only the progress guarantee you can prove, choose reclamation by reader lifetime, know the impossibility walls. Primed once at SessionStart. Grounded in Herlihy's published corpus (consensus hierarchy, linearizability, universal construction, Pass-the-Buck), with the NOT-Herlihy attributions flagged.
1 skill · 1 command · hooks
Routes any decision, action, or iteration through John Boyd's real OODA loop — Observe (measure, don't assume), Orient (the schwerpunkt: reframe against stale models and bias — the highest-leverage step), Decide (pick one lever as a hypothesis), Act (ship it as a test, re-observe), loop fast. Fires a conditional UserPromptSubmit nudge on decide/act/iterate/plan/compete language, primed at SessionStart. Boyd-faithful: not a 4-box speed circle — Orient dominates, and tempo means unpredictable rhythm that outpaces the problem, not raw haste.
1 skill · 1 command · hooks
The sanctioned OS 'motor cortex' for an agent on a Linux box: control systemd services/timers, query journald, read host resources/processes, send desktop notifications, drive D-Bus, and manage power (suspend/reboot/poweroff) — all through structured interfaces (systemctl/loginctl/journalctl/busctl), never raw PID hacks. Pure standard library, zero pip deps. A self-preservation guard refuses severing actions on units the agent depends on (dbus, logind, sshd, network, tailscaled, the session, goosed) unless force=true. Ships the MCP server plus a control-os skill. The system-service counterpart to screen-mcp's GUI control.
2 skills · 1 MCP server
Guards every novelty or superlative claim before it ships: when a turn asserts a first/fastest/novel/record/'nobody has done this', a UserPromptSubmit hook fires the invention discipline — confirmed-limit entry, the removed-constraint test, refute-then-build vs a tuned baseline, provenance search, scope to the measured delta, certify. Primed once at SessionStart. Turns 'we invented X / it's the fastest' from an assertion into a proven-or-scoped claim. Kills overclaim, reinventing prior art, and premature-first-before-refute.
1 skill · 1 command · hooks
Recover Claude Code sessions after an Anthropic API output-classifier false positive (the "cyber-related safeguards" / "appears to violate our Usage Policy" hard failure). A JSON-aware .jsonl log surgeon that surgically removes the triggering message + refusal turn and re-stitches the parentUuid chain so the session resumes cleanly, cleans accumulated refusals across old sessions, and installs a per-repo CLAUDE.md framing guard to stop recurrence.
1 skill · hooks
Eyes and hands on a Linux Wayland desktop: screenshot any monitor and click, type, scroll, drag, and read any visible app over xdg-desktop-portal (RemoteDesktop + ScreenCast), with optional OCR + OmniParser icon grounding. Includes screen_watch (~1 fps human-eye observe) so agents catch thrashing UIs like people do. Pure-Python, CPU-only. GNOME/Wayland only. Ships the MCP server plus a drive-screen skill (locate-ground-act-confirm/watch).
1 skill · 1 MCP server
Talk to any public GitHub repo's auto-generated documentation via Cognition's hosted DeepWiki MCP. Read-only research into codebases without cloning. Note: this plugin wraps a remote MCP server hosted by Cognition AI at mcp.deepwiki.com; 88plug does not operate the underlying service.
1 MCP server
Index a public GitHub repo's DeepWiki (the page the 'Ask DeepWiki' badge links to) hands-free, and do it autonomously after publishing or updating a repo — no human clicking. Encodes the verified deepwiki.com trigger (api.devin.ai/ada/index_public_repo, reCAPTCHA v2-invisible) and the trusted-browser path: drive the real desktop browser (screen-mcp) or run the bundled Playwright helper with a real Chrome profile so the captcha passes silently, then poll the public status API to completed.
1 skill
Operate and observe an interactive full-screen TUI on a REMOTE machine over tmux/screen + SSH by driving it like a human: type with send-keys, screenshot with capture-pane, in a type-wait-screenshot-read loop. For the Claude Code TUI, vim, top, curses installers, REPLs, or any program that needs a real PTY over SSH.
1 skill
Discover, catalog, and rank everything you've built or sketched on a machine via a two-pass parallel read-only sweep: a clustered project catalog plus blind-spot agents (transcripts, other agent CLIs, running services, research artifacts, beyond-home), synthesized into a tiered novelty/leverage ranking with idea/live/dormant tags and evidence-anchored rationale.
1 skill
Fast, token-efficient MCP for SearXNG metasearch. Privacy-respecting search across 70+ engines with stdio + streamable-http transports, Docker, and optional rendered (Playwright) fetch for JS-heavy pages. Self-hostable. The underlying server is also usable independently of Claude Code via `uvx --from git+https://github.com/88plug/searxng-mcp searxng-mcp`.
1 MCP server
Stop suggesting stale package versions from training data. 15 MCP tools over 39 package registries (npm, PyPI, Maven, crates.io, Go, RubyGems, NuGet, Packagist, Docker/GHCR/Quay/GCR, Hex, pub.dev, CRAN, Homebrew, and more): latest-version + package-info lookups, semver compare, install-command generation, batch checks, project dependency scanning, outdated reports, conflict detection, compatibility checks, and a scan -> optimize -> validate -> apply upgrade pipeline that edits manifests with dry-run + automatic backups by default. Local-first, no API key required (an optional GITHUB_TOKEN raises GitHub rate limits).
1 MCP server