Proctor
Test fixture: Proctor loaded with every userConfig option set away from its default.
new
Test fixture: Proctor loaded with every userConfig option set away from its default.
Personal Claude Code plugin marketplace. 10 plugins covering automation, orchestration, code review, debugging, multi-model debate, SEO analysis, philosophical reasoning, hook-enforced development discipline, and a live tracker for obra/superpowers.
Add the marketplace:
/plugin marketplace add nalyk/nalyk-skillsInstall individual plugins:
/plugin install <plugin-name>@nalyk-skillsGates imperative coding tasks (en/ro/ru) into Ralph Loops. A silent-by-default UserPromptSubmit hook flags matches; a deterministic 0-4 scorer routes score >= 3 to /ralph-loop.
/plugin install auto-ralph@nalyk-skillsTriggers: "ralph this", "auto ralph", "loop it" -- or auto-detects bug fixes, features, and refactoring tasks (score >= 3).
Configuration: ~/.claude/auto-ralph.local.md
| Parameter | Default | Description |
|---|---|---|
| max_iterations | 25 | Max Ralph Loop iterations |
| score_threshold | 3 | Min score for activation |
| skip_explore_for_score | 4 | Skip Explore phase at this score |
| default_language | ro | Output language (ro/en/ru) |
| auto_execute | false | Skip confirmation prompt |
| docker_analysis | true | Include Docker context |
Output: Romanian. Input: ro/en/ru/mixed.
Decomposes a task into parallel subagent workstreams, routes them to appropriate agents, and synthesizes results.
/plugin install orchestrator@nalyk-skillsCommands:
| Command | Purpose |
|---|---|
/orchestrate <task> | Full decomposition + parallel execution + synthesis |
/parallel <tasks...> | Quick parallel launch without decomposition |
/plan-only <task> | Preview execution plan without running |
Commands-only — no auto-trigger. The skill activates exclusively via the three commands above.
Three audit frameworks: Steve Jobs (design simplification, 13 questions), George Carlin (BS detection, 13 questions), Vibe (engineering quality, 20 scored metrics).
/plugin install audit-agent@nalyk-skillsCommands:
| Command | Purpose |
|---|---|
/jobs-audit <target> | Design thinking -- simplification, elegance, restraint |
/carlin-audit <target> | BS detection -- hidden agendas, euphemisms, contradictions |
/vibe-audit <target> | Engineering quality -- 20 metrics, 0-5 scale each |
/multi-audit <target> | Run multiple frameworks + cross-reference synthesis |
Triggers: "feature bloat" / "design audit" (Jobs), "BS detector" / "marketing speak" / "corporate speak" (Carlin), "vibe check" / "slop check" / "technical due diligence" (Vibe). Generic phrases like "simplify" and "code review" are intentionally not claimed.
Multi-model adversarial debate. Claude defends a position against external CLI models (agy/Gemini, Codex, Mistral Vibe) in parallel rounds. Produces consensus, tradeoff documents, or ADRs.
Refuses to run with Claude-only. Requires at least 1 external CLI.
/plugin install debate@nalyk-skillsPrerequisites -- at least one:
| CLI | Install | Notes | |
|---|---|---|---|
| agy (Gemini models) | Antigravity CLI v1.0.7+, per vendor docs | pinned to a Gemini model; the legacy gemini CLI is dead/unsupported | |
| Codex | npm i -g @openai/codex | ChatGPT Plus | |
| Mistral Vibe (Devstral) | `curl -LsSf https://mistral.ai/vibe/install.sh \ | bash` | then vibe --setup; free API key at console.mistral.ai |
Commands:
| Command | Purpose |
|---|---|
/debate <topic> | Full adversarial debate |
/debate:doctor | Check CLI availability and auth |
/debate:adr <topic> | Debate with formal Architecture Decision Record output |
Configuration: ~/.claude/debate.local.md
Elite Senior Principal Engineer persona. Four specialized diagnostic protocols for code review, system design, UI analysis, and algorithmic debugging. Romanian-flavored.
/plugin install diagnosticianul@nalyk-skillsProtocols:
| Protocol | Trigger | Function |
|---|---|---|
| protocol-critic | Code snippets, PRs | Forensic code autopsy |
| protocol-architect | "Design a system" | Rigid system planning |
| protocol-visual | UI/CSS/frontend | UI quality enforcement |
| protocol-core | Algorithms, bugs | Surgical debugging |
Triggers: noun-domain signals only (race condition, time complexity, deadlock, database schema, API design, CSS/typography/accessibility). The main persona is invoked explicitly; generic verbs like "fix bug" or "code review" are not claimed.
Philosophical reasoning engine. 63 principles (0-62) from 20 philosophers applied as a decision engine and code review framework. Based on Organon by Lightcyphers SRL.
/plugin install organon@nalyk-skillsCommands:
| Command | Purpose |
|---|---|
/organon | Auto-detect mode and depth from context |
/organon:decide <topic> | Explicit decision analysis |
/organon:review <path> | Philosophical code review |
Depth levels: quick (1 principle), standard (multiple principles), deep (full 22-step protocol + Summa Method objections).
Philosophers: Aristotle, Aquinas, Kant, Machiavelli, Peirce, Plato, Poincare, Popper, Seneca, Stoics, Swinburne, Wittgenstein, plus Leibniz, Boole, Frege, Godel, Turing, Shannon, Church, Marcus Aurelius, Epictetus.
License: CC-BY-SA-4.0
Credits: Original idea and code by Anatolie Golovco.
Deterministic SEO analysis engine. 98 atomic checks across 7 categories, async multi-page crawler, real Core Web Vitals via PageSpeed Insights, internal link graph with PageRank, auto-fix generation, audit history with regression detection.
/plugin install seo-skill@nalyk-skillsPost-install -- Python dependencies (from the installed seo-skill plugin directory):
python3 -m venv .venv
.venv/bin/pip install -r requirements.txtRequirements: Python 3.10+. Optional: PAGESPEED_API_KEY (free, 25K queries/day).
Commands:
| Command | Purpose |
|---|---|
/seo audit <url> | Full site audit with deterministic scoring |
/seo fix <url> | Generate ready-to-apply fixes |
/seo linkgraph <url> | Internal link graph + PageRank |
/seo technical <url> | Technical SEO (9 categories) |
/seo content <url> | E-E-A-T and content quality |
/seo schema <url> | Schema detection, validation, generation |
/seo page <url> | Deep single-page analysis |
/seo history <url> | Audit score history |
/seo monitor setup | Generate CI/CD configs (GitHub Actions, GitLab CI) |
One orchestrator skill plus 7 specialist subagents, CI/CD integration.
Powerline-style status bar for Claude Code. Shows model badge, git status, context window usage, vim mode.
/plugin install statusline@nalyk-skillsRequirements: jq, terminal with Unicode support. Powerline font recommended.
Auto-configures via SessionStart hook only when no statusLine is set; never overwrites an existing one. Restart Claude Code after install.
Hook-enforced development discipline for Claude Code. Skills teach methodology; hooks enforce compliance mechanically (git gates, planning mode, step/time budgets, quiet mode, pre-flight checks, self-diagnosis, autonomy metrics); store + tracing survive compaction.
/plugin install proctor@nalyk-skillsRequirements: Claude Code CLI >= 2.1.260, CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1, a git repository.
Skills (14):
| Skill | Purpose |
|---|---|
using-proctor | Bootstrap -- skill discovery and hook awareness |
brainstorming | Turn ideas into designs (spike/bounded/architectural) |
test-driven-development | Red-green-refactor with git gate enforcement |
systematic-debugging | Root cause before fixes, four-phase investigation |
verification-before-completion | Evidence before claims, mechanically enforced |
subagent-driven-development | Fresh agents per task with dashboard + state persistence |
executing-plans | Inline plan execution with same enforcement as SDD |
writing-plans | Create implementation plans from specs |
requesting-code-review | Dispatch reviewers with proper packages |
receiving-code-review | Evaluate feedback technically |
finishing-a-development-branch | Verify -> present options -> execute -> clean up |
using-git-worktrees | Workspace isolation with branch protection |
dispatching-parallel-agents | Independent concurrent tasks |
writing-skills | TDD applied to skill creation |
Hard gates (5): Test evidence, test freshness, test passing, branch protection, secret/credential detection. Planning mode blocks Write/Edit during design phases.
Soft enforcers (12): Skill watchdog, model selection, fix-round cap, step budget, time budget, context pressure, ruling aggregation, rationalization detection, destructive command warning, diff size awareness, test hint, task advance guidance.
Operator commands (8): proctor: status, proctor: show trace, proctor: check (pre-flight gate status), proctor: diagnose (self-analysis), proctor: tasks N (scope update), proctor: quiet on/off, proctor: allow <branch>, proctor: approve design.
Autonomy features: Failed approach tracking with "DO NOT REDO" compaction injection, SDD session recovery, task completion evidence, structured "Next steps:" on all gate denials, autonomy metrics (gate pass/deny ratio).
Observability: Structured event tracing, cross-session quality metrics, phase lifecycle tracking, real-time SDD progress dashboard.
Configuration (6 options): Protected branches, test freshness window, skill watchdog threshold, fix-round cap, step budget per task, time budget per task — all configurable via the plugin settings UI.
Without CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1, the skills still work as prose guidance but hooks do not fire.
A live tracker for obra/superpowers inside Claude Code: a stage rail in the band above the prompt (brainstorm → spec → plan → execute → review → finish), a seven-tab dashboard pane, toasts when work lands, and resume detection for unfinished plans. An observer first: by default it never blocks, rewrites or delays anything, and never writes inside superpowers' own folders.
/plugin install superpowers-tracker@nalyk-skills --scope userInstall it at user scope: it stays dormant in projects that don't use superpowers and fires on its own where they do.
Requirements: Claude Code with mods (function hooks), built and tested on 2.1.289. Built against superpowers 6.4.2; every superpowers format it reads lives in hooks/patterns.ts.
What it shows:
| Surface | Content |
|---|---|
| Band | Brand capsule, path tag (ARCH/BOUNDED/SPIKE), the arc rail with a sub-cell progress bar, ⚖ rulings / ⌁ agents / ⚠ warnings pills; a second row with the running subagent, fix-round pips, elapsed time, the TDD trail and a context sparkline |
Pane (/sp) | Flow (subway map), Plan (tasks + Gantt strip), Agents (swimlanes), Ledger (rulings, parked findings, deferred minors), Tests (RED→GREEN), Workspace, Warnings |
| Resume | On every start it scans this checkout and all git worktrees for plans in progress, plans not executed, specs without a plan, a running visual companion, and ledgers still moving in another session; offers the resume text as ghost text and fills it on r (never submits) |
Modes: observe (default, records only), assist (adds a ≤ 3-line ledger note after compaction or a re-dispatch), strict (also denies re-dispatching a completed task, parallel implementers, SDD subagents with no model, and source writes before brainstorming approval).
Commands: /sp [1-7], /sp status, /sp ledger, /sp export, /sp ack, /sp mode, /sp band, /sp config, /sp reset, /sp resume [n], /sp dismiss [n] [--forget], /sp scan, /sp demo, /sp theme, /sp diagnose.
Design: one hue per stage, blended in OKLCH; every text token at ≥ 4.5:1 in dark, light and daltonized themes; single-width glyphs only; motion only while work runs, none when idle or under NO_COLOR. Desktop gets an interactive SVG rail; VS Code and mobile a Box/Text version.
61 engine tests (claude plugin test), run by make test.
See individual plugin READMEs in plugins/<name>/ for full documentation.
MIT (unless noted otherwise per plugin)
Not captured.