ABVX Lab logo ABVX Lab
skill catalog
ABVX Agent Skills logo

ABVX tool

ABVX skill catalog

Find the right skill fast, including bounded Planner/Reviewer/Executor orchestration, opt-in project guardrails, L1/L2 loop contracts, product context, design-taste routing, review lenses, skill-quality evolution, and loop readiness.

GitHub PyPI LoopOps v0.11.0
Primary role Searchable skill inventory
Source of truth

Generated from abvx-agent-skills main. Lab is the published surface, not a manual fork.

What it keeps visible

Category, short description, starter guidance, source links, delivery gates, bounded orchestration, opt-in guardrails, context/tool-contract review, marketing/editorial taste review, product-UI evidence routing, eval guidance, MCP/CLI routing, and next actions.

Why this version exists

The old page acted like a product pitch. This one is the working catalog, including delivery discipline patterns and the access/execution/discipline layering behind ABVX skills.

Search

This page pulls the generated catalog from the repository and keeps the Lab URL stable: /tools/abvx-agent-skills/.

Workflow layers

  • Access MCP and integrations connect external systems without putting every service contract into the prompt.
  • Execution CLI tools, scripts, CI, and local commands do deterministic work that can be logged and reviewed.
  • Discipline Skills decide when access and execution are allowed, which proof gates apply, and when a repeated workflow should become a reusable capability.
  • Project hook guardrails project-hook-guardrails and hook-rule-fixtures keep data-boundary and promotion checks opt-in, per-project, fixture-tested, and explicit about warn versus block. They do not install global observer or session hooks.
  • Review lenses assumption-excavation, agent-tool-contract-review, agent-operations-contract, frontend-taste-layer, bounded-evaluation, loop-readiness-review, and loop-hardening-contract turn agent, design, and pipeline patterns into compact ABVX gates. External adaptations receive source-linked deltas; named agents receive explicit authority cards, operation and decision receipts, trust-graded memory scopes, and provider/tool evidence.
  • Exploration rabbithole-doc-exploration uses a local canvas for human-selected questions over AGENTS.md, skill docs, SET plans, and repomaps.

Top starts

  • idea-to-ship-gates Route vague product work through intent, spec, slices, proof, review, and release.
  • social-publishing-gate Draft, audit, approve, publish, and monitor social content without unsafe automation.
  • minimal-diff-builder Shortest correct implementation path for real coding work.
  • diagnose Debug around one signal, one hypothesis ladder, and narrow verification.
  • assumption-excavation Surface hidden assumptions before implementation or runner handoff.
  • pipeline-readiness-gate Select a compact pre/post/ship gate without adopting a full pipeline runtime.
  • confidence-fragility-review Check whether confident claims are backed by evidence before ship.
  • knowledge-base-enrichment Turn useful findings into auditable, project-scoped Markdown knowledge without unsafe memory writes.
  • context-degradation-review Check for context poisoning, lost-in-middle risk, distraction, and instruction clashes.
  • loop-readiness-review Decide whether a recurring workflow is ready for bounded loop execution.
  • bounded-orchestration-contract Gate plan review, stable findings, disjoint executor packets, route evidence, and mandatory root verification.
  • loop-hardening-contract Use measured harness stripping, immutable runtime-path sprint predicates, and broken-window revalidation without automatic revert.
  • agent-operations-contract Review concrete agent authority, async and revalidated decision receipts, trust-graded scoped memory, public/private state boundaries, provider/tool availability, and human approval boundaries without installing a runtime.
  • project-hook-guardrails + hook-rule-fixtures Add one project-local warning or named-verification gate, then prove it against synthetic fixtures before rollout.
  • rtk-assisted-shell Best early win when shell output is burning your session budget.
  • rabbithole-doc-exploration Open dense agent docs in a local Rabbithole canvas before changing durable contracts.
  • frontend-product-builder Product-aware frontend work instead of generic component churn.
  • handoff Compact continuation briefs when work spans multiple sessions or operators.

Starter packs and next actions

  • Solo dev baseline minimal-diff-builder, diagnose, rtk-assisted-shell, token-efficient-execution
  • Frontend stack frontend-product-builder, design-critique-polish, browser-verification, designmd-brand-kit
  • Delivery discipline idea-to-ship-gates, plan-to-issues, delivery-preflight-gate, delivery-baseline-audit
  • Content distribution social-publishing-gate, evidence-ledger-research, private-vs-publishable-skill-audit, agent-learning-layer-triage
  • Agent improvement agent-friction-ledger, durable-context-maintenance, token-usage-audit, skillopt-evolve-skills for staged rollout evidence, bounded candidate edits, held-out validation, rejected-edit memory, and accepted best_skill.md export
  • Product and growth review product-context, bounded-growth-loop, bounded-evaluation, loop-readiness-review
  • Team standardization project-context-bootstrap, durable-context-maintenance, brief-first-execution, handoff
  • Knowledge and memory knowledge-base-enrichment, filesystem-context-discipline, durable-context-maintenance, context-degradation-review
  • Loop readiness loop-readiness-review, loop-hardening-contract, bounded-evaluation, bug-evidence-protocol
  • Project guardrails workflow-policy-layering, project-hook-guardrails, hook-rule-fixtures, delivery-preflight-gate

Install

python -m pip install abvx-agent-skills
abvx-skills list
abvx-skills install dynamic-workflow-packets

Prefer this path when you want packaged install, validation, and CLI support.

Raw bundle

git clone https://github.com/markoblogo/abvx-agent-skills
cp -R abvx-agent-skills/skills/* ~/.codex/skills/

Use the raw bundle when testing local edits or when packaging is not the point.