Skip to content
View waitdeadai's full-sized avatar
1 follower · 1 following

Achievements

Achievement: Pull Sharkx2Achievement: Pair ExtraordinaireAchievement: Quickdraw

Achievements

Achievement: Pull Sharkx2Achievement: Pair ExtraordinaireAchievement: Quickdraw

Block or report waitdeadai

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
waitdeadai/README.md

Fernando Lazzarin

Founder, WAITDEAD · vertical intelligence studio · Mendoza, Argentina

Shipping doesn't sleep. Decisions, not dashboards.

What I'm building

  • WAITDEAD — vertical intelligence studio. Eight catalogs, one engine. PriceIntel live for retailers and distributors; LeadRadar, CompetiSignal, RateScope, EstateLens, HireSignal, RegWatch, SupplyPulse on demand (7–14 day build). Free 48-hour sample brief, USD 600–1,500/mo per product, cancel any time.
  • REVCLI — governed AI workflow platform with paying customers in production. Human approval gates, audit trails, policy engine.

Just shipped

no-vibes empirical baseline against the MAST (Multi-Agent System failure Taxonomy, Cemri et al., NeurIPS 2025 Datasets & Benchmarks spotlight) MAD dataset:

  • F1 0.815 (95% CI [0.615, 0.941]) on the n=19 human-labelled subset against MAST mode 3.3 ("No or Incorrect Verification")
  • Fleiss κ 1.000 inter-annotator agreement on mode 3.3 specifically (paper's overall κ=0.88 is on a different 150-trace taxonomy-development set)
  • Per-MAS breakdown: result is not driven by a single framework
  • Productized as crewai-no-vibes — pip install crewai-no-vibes, pure-Python CrewAI Task guardrail, zero runtime deps

Full empirical write-up with bootstrap methodology, per-MAS breakdown, and per-mode Fleiss kappa: evaluation/MAST-RESULTS.md.

The discipline that makes this credible: of 13 hooks conceptually mapped to MAST modes, only no-vibes produces measurable F1 > 0 at the trace-level baseline. The other 12 are documented honestly as "conceptually mapped, no measured signal yet" rather than overclaimed. Honest demotion compounds.

Open-source highlights

Repo What it does
llm-dark-patterns Apache-2.0 28-hook suite for runtime enforcement of LLM dark patterns at the Claude Code Stop event. Out-of-band bash + jq judge, deterministic; the model that produced the closeout cannot rewrite the verdict from inside its own output.
agent-closeout-bench Rust YAML rule-pack engine + 175 fixtures + 73 tests. Powers the MAST-EVAL + DarkBench v1.5-rust head-to-head.
crewai-no-vibes Pure-Python port of the evidence_claims rule pack as a CrewAI Task.guardrails function. Apache-2.0, zero dependencies, on PyPI.
minmaxing Production Claude Code harness — /opusworkflow, governed model profiles, 5-tier memory, parallel orchestration.
claude-plugins Self-hosted Claude Code plugin marketplace. Bypasses the stalled Anthropic community pipeline.
no-vibes, no-sycophancy, no-cliffhanger Standalone single-file Stop hooks for users who want one hook without the suite.

Research anchors

Where to find me

8+ years RevOps · 50+ audits · 13 industries · 9 CRM platforms. Operated from Mendoza, Argentina under SmoothRevenue LLC (Wyoming).

Pinned Loading

  1. claudemax claudemax Public

    Anthropic-native power-user harness — spec-driven multispec, Mode A/B parallelism, Claude Max plan-aware, bundled dark-patterns, remote-from-phone via Tailscale + tmux + ntfy

    TypeScript

  2. llm-dark-patterns llm-dark-patterns Public

    Umbrella for the LLM Dark Patterns Hooks suite — single-purpose Claude Code Stop hooks that suppress sycophancy, paternalism, false-success, permission-loops, training-cutoff confidence at the text…

    Shell 10

  3. no-vibes no-vibes Public

    Claude Code Stop hook that blocks false-success closeouts. Single bash file, jq-only dep, drop-in. Receipts inside.

    Shell 1

  4. agent-closeout-bench agent-closeout-bench Public

    Deterministic closeout physics engine and benchmark for agentic coding assistant dark-pattern detection

    Python 1

  5. crewai-no-vibes crewai-no-vibes Public

    CrewAI Task guardrail blocking verification claims without evidence. Pure-Python port of the no-vibes / MAST mode 3.3 detector. F1 0.815 (95% CI [0.615, 0.941]) on the released MAD human-labelled set.

    Python

  6. minmaxing minmaxing Public

    The ultimate MiniMax 2.7 Highspeed Claude Code harness - 100 TPS, 204K context, GStack skills, 5-tier memory, 10x productivity.

    Shell 5