# Agent Capability Catalogue (descriptor, not a brand) Source-backed prompts and skills for autonomous agents, indexed by the problem they solve. This file is a convenience index for language-model and tool consumers; it is not claimed as a universal protocol. ## Discovery - agent guide: https://agents.proofandstate.com/agents - agent discovery text: https://agents.proofandstate.com/agents.txt - canonical catalogue: https://agents.proofandstate.com/catalog.json - per-product metadata: https://agents.proofandstate.com/raw/products/{slug}.md ## Catalogue - schemaVersion: 1.0.0 - products: 16 - types: prompt=11, skill=5 ## Data policy - Public records are source-backed metadata. - No prices, ratings, reviews, compatibility or sales figures are invented. - Full PROMPT.md and SKILL.md payload bodies are withheld. - Missing fields are omitted rather than fabricated. ## Agent interface - HTTP discovery is available now through the endpoints above. - CLI search/show/sample/buy/install/update is roadmap only. - MCP is out of scope. ## Categories - Agent Architecture (agent-architecture) - Agent Governance (agent-governance) - Agent Operations (agent-operations) - Agent Security (agent-security) - Agent Workflows (agent-workflows) - Developer Tooling (developer-tooling) - Developer Workflows (developer-workflows) - Frontend Verification (frontend-verification) - Marketing Systems (marketing-systems) - Production Operations (production-operations) - Production Reliability (production-reliability) - Publication Reliability (publication-reliability) ## Products - Autonomous Coding Workflow with a Work Ledger [prompt] — Build a resumable coding-agent workflow that selects bounded work from repository evidence, persists decisions in a durable ledger, verifies work before completion, and stops at explicit approval boundaries. (https://agents.proofandstate.com/raw/products/autonomous-coding-workflow.md) - Bounded Google Docs Read Recovery [skill] — Diagnose Google Docs reader failures across network routes, OAuth refresh, Drive metadata and bounded document responses, then verify a compact-response repair without weakening document integrity checks. (https://agents.proofandstate.com/raw/products/bounded-google-docs-read-recovery.md) - Capability-Gap Learning System [prompt] — Build a governed system that classifies why agent work is blocked, separates missing capability from missing permission, proposes the smallest justified improvement, validates it, and retries the original task without self-granting authority. (https://agents.proofandstate.com/raw/products/capability-gap-learning-system.md) - Continuous Autonomous Business Operations Graph [prompt] — Build a durable, change-aware business operating loop that discovers, prioritises, executes, verifies, and replans useful work without relying on a fresh user prompt after each task. (https://agents.proofandstate.com/raw/products/continuous-business-operations-graph.md) - Deterministic Social Publication Pipeline [prompt] — Build an auditable social publishing pipeline with immutable content specifications, durable slot ownership, one bounded provider write, official readback, reconciliation, and evidence-backed terminal states. (https://agents.proofandstate.com/raw/products/deterministic-social-publication.md) - Evidence-Backed Open-Source Contribution [skill] — Take a bounded open-source issue from selection to maintainer-ready submission using contributor-rule discovery, collision checks, exact-base branching, red/green proof, repository-native validation, and honest CI/review reporting. (https://agents.proofandstate.com/raw/products/evidence-backed-open-source-contribution.md) - Evidence-First Live Diagnostic and Repair [prompt] — Diagnose and repair external-write failures without duplicate side effects by ranking evidence, separating ambiguous outcomes, and requiring authoritative provider readback before success. (https://agents.proofandstate.com/raw/products/evidence-first-live-diagnostic-repair.md) - Graph-Native Agent System Migration [prompt] — Migrate an existing automation or agent system from loosely connected jobs and schedulers into a durable graph runtime with checkpoints, explicit authority, idempotent effects, recovery, and provider-side verification. (https://agents.proofandstate.com/raw/products/graph-native-agent-system-migration.md) - Live Dashboard Update Verification [skill] — Verify live dashboard updates from observation to visible state, including shared DOM ownership, complete initial snapshots, successive events across views, freshness, and retained reading state under the production content security policy. (https://agents.proofandstate.com/raw/products/live-dashboard-update-verification.md) - Local-First Verification CLI [prompt] — Build a stack-aware, read-only CLI that classifies verified facts, warnings, failures, skipped checks, and unverified production claims before developers trust AI-generated changes. (https://agents.proofandstate.com/raw/products/local-first-verification-cli.md) - Production Agent Operating Files [prompt] — Design and implement the complete operating-file pack and runtime contract for a specialised agent, aligning role, scope, policy, tools, memory, machine-enforced permissions, tests, runtime health, and operator evidence. (https://agents.proofandstate.com/raw/products/production-agent-operating-files.md) - Published Package Contract Verification [skill] — Verify and repair the package users actually install by testing built exports, declaration graphs, dependency boundaries, and consumer compilation instead of trusting source-only checks. (https://agents.proofandstate.com/raw/products/published-package-contract-verification.md) - Read-Only Production System Reconnaissance [prompt] — Map a live software system without mutating source, services, schedules, databases, credentials, providers, or runtime state, while reconciling documentation against observed runtime truth. (https://agents.proofandstate.com/raw/products/read-only-production-reconnaissance.md) - Safe Scheduled Runtime Upgrade [skill] — Upgrade scheduled runtimes through an exact target, verified job draining, a fresh guarded review, recoverable timer restoration, and separate process-readiness and operational-completion checks. (https://agents.proofandstate.com/raw/products/safe-scheduled-runtime-upgrade.md) - Self-Identifying Product Campaign [prompt] — Build evidence-backed campaigns that help the right audience recognise an observable current problem before presenting a relevant product, service, or diagnostic next step. (https://agents.proofandstate.com/raw/products/self-identifying-product-campaign.md) - Signed Agent Action Receipts [prompt] — Build an evidence-bound execution system where one bounded consequential action requires transaction-specific approval, executes exactly once, is independently verified, and produces an offline-verifiable signed receipt. (https://agents.proofandstate.com/raw/products/signed-agent-action-receipts.md)