tangle-network

Adopt agent-eval: run the real agent, capture records, calibrate cases, and control promotion.

Integrate or audit Tangle runtime, evaluation, knowledge, profiles, and candidate activation.

Build or migrate an agent product with agent-app modules, billing, integrations, and evals.

Build Tangle Blueprint flows for jobs, operators, provisioning, auth, and shared UI.

Review a plan by expanding scope, maximizing rigor in scope, or reducing to essentials.

Build a sandbox Blueprint covering isolation, jobs, APIs, auth, secrets, and lifecycle.

Build on sandbox with sessions, browser streaming, scoped tokens, usage, billing, and edge.

Apply Tangle design tokens and components to Sandbox product UI and responsive states.

Wire product measurement and improvement to Agent Eval campaigns, official optimizers, and explicit release decisions.

Optimize one production agent surface with Agent Eval methods, separate data, bounded spend, and measured promotion.

The user-facing controller for the Improve button. Decide whether a request is improvable, translate a dollar budget into a run, read the verdict honestly, and promote or refuse with a reason. Never promise a lift you cannot measure.

Prove a measurement is sound BEFORE spending money optimizing against it. The gate that decides whether an Improve run is allowed to start, and whether its result is allowed to be believed. Refuse metrics whose noise exceeds the effect, that have no held-out…

How every skill in the Improve family stays agentic and general instead of rotting into a brittle rulebook. A skill is a measured hypothesis — a few human-owned invariants plus a wide loop-owned judgment surface that improves from outcome data via its own…

Build a measurement that scores an agent's REAL deliverable — not a proxy — for a product you've never seen before. Use when scaffolding or repairing the eval an Improve loop optimizes against. Get this wrong and every downstream optimization perfects a…

When a product has NO improvement infrastructure yet, build it for real — elicit the RIGHT target, ground the measurement in external truth, and construct a validated harness (often via a delegated agent-runtime build loop) BEFORE any optimization spend. The…

Batch mechanical tool work as one program so loops and intermediates stay out of context.

Author runGraph programs from AgentProfiles and versioned prompt directives.

Choose and compose current runtime, eval, knowledge, and interface APIs before adding wrappers.

Write a custom control policy only when current runtime composition APIs cannot express it.

Create one pinned fresh-API coding case that passes execution and defeats a no-search baseline.

Drive worker agents with explicit profiles, corrections, budgets, and checked delivery.

Maintain agent-eval cases, judges, records, traces, campaigns, comparisons, and releases.

Parse classic SWE-agent .traj trajectories: one step per entry of the trajectory array, in order, matching the CodeTraceBench annotation convention. No upstream CodeTracer commit ships a SWE-agent parser.

Parse OpenHands trajectories published as per-call LiteLLM completion logs (<model>-<timestamp>.json files with messages/response/kwargs/cost), the layout of CodeTraceBench's swe_raw OpenHands trials. Reads the call file whose messages hold the most tool-call…

Parse OpenHands session-event trajectories at the CodeTraceBench annotation granularity: one step per action event with a cause-paired observation (run, run_ipython, read, edit, think, recall). Outranks the seed openhands skill, which counts only…

Parse Terminus2 trajectories at command granularity: one step per entry of each episode response's commands array, matching the CodeTraceBench annotation convention. Outranks the seed terminus2 skill, whose episode-level steps do not align with the…

Validate Tangle blog MDX before publishing. Use to check frontmatter, answer quality, internal/external links, proof, decisions, image references, AEO/SEO gates, and repeated AI-writing patterns.

Audit, rewrite, or batch-rewrite Tangle website blog posts in src/content/blog. Use for existing MDX posts that need stronger Tangle voice, SEO/AEO structure, answer capsules, proof blocks, internal links, FAQ, publication-ready technical clarity, or a…

Create new Tangle agent-intent blog posts from query briefs, gists, outlines, or product specs. Use for posts targeting AI agents, AEO, SEO, Sandbox, Browser Agent, Router, x402, Blueprints, TEE attestation, manifests, OpenAPI, llms.txt, or package discovery.

Closed-loop design improvement using bad design-audit. Audits a running app, reads findings, applies fixes to actual source code, re-audits to verify improvement, iterates until converged. Use when the user says 'improve the design', 'design audit and fix',…

Use when configuring browser stealth profiles, anti-bot evasion, WebDriver detection suppression, JA3 fingerprinting, or CDP leak patches.

Inspect real agent workflows with the published Traces CLI and export cited local findings.

Write and calibrate a custom agent-eval trace analyst on real and labeled workflows.

Build, evaluate, and improve source-backed knowledge, retrieval, and memory systems.

Run the `bad` CLI (Browser Agent Driver) against the local frontend with a web3 wallet (MetaMask) connected. Use this skill when the user wants to browse, test, or interact with the local UI at localhost:1338 using a connected wallet via `bad`. Triggers on:…