Ken Huang

@kenhuangus·29 skills·1 follower

Writes Agentic AI on Substack. Agentic AI security, harness engineering and agent skills; OWASP Agentic Skills Top 10. DistributedApps.AI.

Agentic AI (Substack)GitHub OWASP Agentic Skills Top 10

Skills29

verifier-subagentCheck finished work with a fresh-context subagent instead of self-critique when using Claude Fable 5. Use when designing multi-agent harnesses, verifying long-run output, or any time self-review keeps missing its own mistakes -- the maker should not grade its own work.Free

scope-fenceStop Claude Fable 5 from taking actions nobody asked for -- applying a fix when only asked to diagnose, restarting a service, creating a backup branch, editing a config on a hunch. Use for ops/debugging sessions, production-adjacent work, and shared environments where the cost of an unrequested change is high.Free

refusal-aware-callingWrite and review Claude Fable 5 API calls and prompts so a safety-classifier refusal degrades gracefully instead of dead-ending. Use whenever adding a new claude-fable-5 call, reviewing security/red-team/bio/ML-adjacent workloads that are refusal-prone, or debugging a request that silently returned no content.Free

reader-first-summaryMake Claude Fable 5's final report readable to someone who saw none of the work. Use for long agentic sessions, overnight runs, and multi-tool tasks whenever summaries come back as arrow-chain shorthand, invented abbreviations, or references to reasoning the reader never saw.Free

no-gold-platingKeep Claude Fable 5's diffs mapped one-to-one to the request -- no unrequested refactors, speculative abstractions, or defensive error handling for states that can't occur. Use in code review and any coding session where the diff comes back larger than the ask, especially at high or xhigh effort.Free

mythos-scope-gateUse before any approved Mythos-style defensive cybersecurity run, or when reviewing a Fable/Mythos security workflow. Requires authorization, bounded targets, private evidence handling, independent validation, and a patch/report owner before the model explores vulnerabilities.Free

lesson-memoryMaintain a file-based lesson memory that Claude Fable 5 reads and updates across sessions. Use for recurring agents (daily jobs, long projects, team assistants) where the same correction keeps being re-made, or when bootstrapping a new agent from past session history.Free

evidence-bound-progressBind every progress claim Claude Fable 5 makes during a long or autonomous run to an actual tool result from that session. Use for multi-hour agents, overnight pipelines, CI runs, or any workflow where a status update has previously claimed work that wasn't actually done.Free

effort-routerPick and re-evaluate the output_config.effort level for a Claude Fable 5 workload. Use when designing a pipeline that mixes routine and hard tasks, when a call feels slow or expensive for what it's doing, or when porting an Opus 4.7/4.8 harness that defaulted to xhigh.Free

autonomy-contractKeep Claude Fable 5 running to completion in unattended pipelines -- no stalling on a stated intent without the matching tool call, no mid-run permission questions, no self-imposed session splits from context-budget worry. Use for overnight runs, scheduled jobs, and CI agents where nobody is watching to type "continue".Free

act-on-sufficiencyStop Claude Fable 5 from over-planning, re-deriving settled facts, or listing options it won't pursue. Use in interactive sessions where turns feel slow on simple asks, or agentic sessions at high/xhigh effort where the model gathers more context than the task needs.Free

verify-fixIndependently retests a remediated finding after deploy for the Defense Factory Verified remediation phase. Use when a fix has shipped; a merged PR alone is never sufficient to mark remediation complete.Free

triage-validate-findingDynamically reproduces finding candidates with authorization and writes validated findings for the Defense Factory Dynamic validation phase. Static-only results stay candidates; refuse unauthorized Red testing.Free

threat-modelProduces a concise threat model for in-scope assets during the Defense Factory prepare phase. Use after SECURITY.md exists and before or alongside inventory to prioritize discovery and validation effort.Free

security-scanRuns authorized defensive discovery across inventoried assets and writes finding candidates for the Defense Factory Discovery phase. Use for baseline or periodic scans, not for unauthorized offensive testing.Free

security-diff-scanPerforms change-scoped defensive discovery on a diff or PR for the Defense Factory Discovery phase. Use on pull requests or local uncommitted changes instead of a full-fleet scan.Free

route-ownershipAssigns accountable owners and SLAs to validated findings for the Defense Factory Ownership phase. Use after validation and deduplication when fix responsibility must be explicit.Free

fix-findingImplements a verified-ready code or config fix for an owned finding in the Defense Factory Verified remediation phase. Use after ownership assignment; does not mark the finding verified—that requires verify-fix after deploy.Free

define-security-policyCreates or updates SECURITY.md and authorization scope for a repo in the Defense Factory prepare phase. Use when starting a defensive assessment, before inventory or scans, or when SLA/ownership rules are missing.Free

dedupe-findingsClusters duplicate and near-duplicate security findings after Discovery or validation in the Defense Factory Dynamic validation phase. Use before ownership assignment to reduce noise and link duplicates to a canonical finding.Free

build-inventoryEnumerates services, repos, APIs, and related assets into schema-valid inventory files for the Defense Factory Inventory phase. Use when starting discovery or when asset ownership is unknown.Free

sentinelc-incident-responseRun SentinelC Windows malware, spyware, keylogger, rootkit, persistence, Defender tamper, offline-like VSS, and cleanup-verification workflows. Use when the user asks to run SentinelC, analyze SentinelC JSON findings, investigate malware leads, remove risky Defender exclusions, disable unwanted persistence, quarantine suspicious files, or verify a cleanup.Free

paa10-guardGuard a personal AI agent against the OWASP Top 10 for Personal AI Agents (PAA10). Use before any action that sends, posts, shares, buys, pays, transfers, deletes, changes account or recovery settings, enters credentials, downloads or runs something, records people, or stores lasting memory; when the owner says stop or no; and when a web page, email, or another agent asks the agent to act.Free

codex-cross-tool-reviewerUse when Codex reviews Claude Code-authored or other-agent-authored changes and verifies ownership claims before integration.Free

compound-orchestratorUse when setting up or enforcing a compound engineering loop across coding, writing, research, or documentation projects with Codex, Claude Code, and other agent runtimes.Free

super-orchestratorOrchestrates complex, long-running, multi-stage engineering and research tasks by dynamically decomposing goals, provisioning isolated subagent worktrees, steering parallel workers asynchronously, and synthesizing verified deliverables.Free

work-hard-protocolOperating protocol for ambitious agent work across writing, coding, research, and general task completion. Adapted from Felix Rieseberg's tutorial "How to be more ambitious: A tutorial for humans with models" (felixrieseberg.com, Oct 2026). Load this skill when baseline execution is not enough and the work should be pushed toward the outcome, the expert standard, and outside disciplines. Free

harness2Run Google's Harness² recursive harness optimization on a custom agent domain by scaffolding a BenchmarkAdapter, validating the c1 to c8 harness tree with Harness²'s own validator, and running a fail-closed Propose-Probe-Compose loop with a frozen model. Use when Ken asks to optimize, evolve, or self-improve an agent harness with Harness², build a Harness² adapter, or run a Harness² demo or sweep.Free

greenfield-software-factoryRun an autonomous greenfield software factory for from-scratch, end-to-end delivery in a blank workspace or explicitly designated new-project directory: specification, architecture, vertical-slice implementation, independent validation, and review-ready handoff. Use only when the user asks for a complete new app, website, service, API, CLI, library, mobile/desktop product, data/ML system, or full-stack product and wants factory-style autonomous execution. Do not use for scaffold/example-only requests or when the requested work modifies an existing product or repository, including isolated features, bug fixes, refactors, review-only tasks, CI repair, maintenance, or deployment-only work.Free