القائمة
Implement OpenAI Harness Engineering practices in any repository — AGENTS.md, PLANS.md, deterministic smoke/test/lint harness commands, strict architecture boundaries, observability from day 1, and entropy-control audits for reliable autonomous agent runs.
| name | harness-engineering-playbook |
| description | Implement OpenAI Harness Engineering practices in any repository — AGENTS.md, PLANS.md, deterministic smoke/test/lint harness commands, strict architecture boundaries, observability from day 1, and entropy-control audits for reliable autonomous agent runs. |
A skills.sh-compatible skill that operationalizes the practices from OpenAI's Harness Engineering guide. Use it to set up or refactor agent-first workflows so that autonomous runs are repeatable, observable, and safe.
npx skills add broomva/skills --skill harness-engineering-playbook
AGENTS.md, PLANS.md, docs/ARCHITECTURE.md, docs/OBSERVABILITY.md, Makefile.harness, and CI workflows.make smoke, make check, make ci so agents can run them reliably.audit — treat any MISSING or FAIL as blocking.# Interactive wizard (recommended)
python3 .agents/skills/harness-engineering-playbook/scripts/harness_wizard.py init <repo-path> --profile control
# Shell fallback
./scripts/bootstrap_harness.sh <repo-path>
# Audit
python3 .agents/skills/harness-engineering-playbook/scripts/harness_wizard.py audit <repo-path>
| Profile | Scope |
|---|---|
baseline | Core harness artifacts only |
control | Baseline + control-system primitives |
full | Control + entropy controls, nightly audit, CI |
OpenAI Harness Engineering guide: https://openai.com/index/harness-engineering/
MIT
revenuecast — turn a real-world capability into a self-demonstrating, high-throughput generative-AI revenue engine (the "Kleos" method). It is "/skillify for revenue": skillify turns a workflow into a tested skill; revenuecast turns a capability into a revenue engine whose own output IS the advertisement. The bstack-native composition of the 2026 "show-then-sell-the-system" creator loop (realosias, aivideoskool, GenHQ): Brand-Lock -> Show -> Distribute -> Hook -> Sell -> Moat, where the showcased output's desirability + accessibility-via-AI creates demand for the method, and you monetize the reproducible system. Composes content-engine (the factory), content-creation, blog-post, seo-llmeo, arcan-glass, social-intelligence, strategy-skills, and symphony/arcan (autonomous runtime). Its deterministic core (scripts/revenuecast_check.py) gates an engine-instance manifest on the design canon — own-the-audience, a real moat (not leakable prompts), the compliance/survival pillar (FTC v. Air AI / EU AI Act Art.50 / NO
Skillify-as-a-verb — distill a working session (or a pointed-at chat history) into a permanent, TESTED, registered skill at the end of a workflow. The bstack-native composition of Garry Tan's 10-step "skillify it": look-back extraction → CreateSkill scaffold → latent/deterministic split → unit tests → resolver-eval (role-x.py eval) → script-test gate (bstack skills audit --require-tests) → P20 cross-review → bookkeeping file. Composes existing primitives; reimplements nothing. The deterministic core (scripts/skillify_check.py) makes "a feature that doesn't pass all ten is not a skill" machine-checkable. USE WHEN: "skillify it", "skillify this", "package this as a skill", "distill this into a skill", "make this a skill", "turn this into a skill", or at the end of an ad-hoc workflow that worked and should become permanent. NOT FOR: ingesting an external artifact (use /checkit); retrospective "what have I done repeatedly" discovery alone (use the look-back lens); a one-off task with no reusable procedure.
Fresh-session handoff doc drafting. Produces a stable, single-file human-readable narrative state for the NEXT agent context (fresh session, after `/clear`, after persist iteration, after a tab close). The artifact lives at `docs/handoffs/YYYY-MM-DD-<arc>.md` and follows a stable shape: TL;DR + State-of-the-world (P15 snapshot) + What-was-delivered (PR table with SHAs) + First action + Pickup state. Distinct from P12 persist's `PROMPT.md` (machine-state for cross-context loop) and the P1 Bridge session log (raw transcript) — the handoff is the narrative bridge a human reads in ten seconds and a fresh agent reads in thirty. Use when: (1) ending a substantive session that another agent will continue, (2) preparing a fresh-session pickup point mid-arc, (3) needing to compress a multi-PR arc into a single resumable document, (4) the user says "write a handoff" / "fresh-session handoff" / "let me come back to this tomorrow". Triggers on "handoff", "fresh-session", "fresh session", "pickup", "where we are", "leave
Investment management skill — portfolio construction, analysis, and execution. Compounds finance-substrate (accounting/tax) + wealth-management (projections/goals) into a full financial framework. Covers traditional investing (stocks, ETFs, bonds), alternatives (crypto, prediction markets, real estate, VC), quantitative analytics (factor models, Monte Carlo, optimization), and platform integration (Alpaca, Coinbase, Polymarket, agent-browser for Colombian platforms). Embodies philosophies from Buffett, Dalio, Bogle, Munger, and Marks. Use when: (1) building or analyzing a portfolio, (2) screening stocks/ETFs/crypto, (3) running backtests or factor analysis, (4) executing trades or rebalancing, (5) tracking investments across platforms, (6) researching market data or fundamentals, (7) making investment decisions with structured frameworks. Triggers on 'investment', 'portfolio', 'stocks', 'ETF', 'bonds', 'crypto', 'trading', 'backtest', 'factor model', 'rebalance', 'Polymarket', 'Alpaca', 'asset allocation'.
Estimate optimal waiting time for Alkosto's "every 25/50 customers" promotion using either checkout-flow observations or winner announcement timestamps. Use when the user asks how long to wait, wants a probability-based cutoff, or needs a fast in-store decision rule with uncertainty handling.
Agentic robotics with CaP-X — LLM-driven robot manipulation via code generation. Use when: (1) Setting up CaP-X / CaP-Gym environments for robot manipulation benchmarks, (2) Running CaP-Bench evaluations across LLMs/VLMs on robotic tasks, (3) Building or extending CaP-Agent0 agentic harnesses (skill libraries, visual differencing, parallel reasoning), (4) Training robot coding agents with CaP-RL (GRPO on code generation), (5) Developing perception APIs (SAM3, Molmo, depth, point clouds) or control APIs (IK solvers, grasp planners), (6) Sim-to-real transfer for Franka Panda, R1Pro humanoid, or other robot platforms, (7) Designing auto-synthesized skill libraries for physical manipulation (Voyager-style), (8) Integrating agentic robotics with Life Agent OS (Arcan orchestration, Spaces agent networking, Lago persistence), (9) Any task involving LLM-based robot control, manipulation benchmarks, robotic code synthesis, or embodied AI agents.