Skip to content

Instantly share code, notes, and snippets.

Show Gist options
  • Select an option

  • Save magnus919/97af85cbeb2c1a37c0ed7c274fcc5f52 to your computer and use it in GitHub Desktop.

Select an option

Save magnus919/97af85cbeb2c1a37c0ed7c274fcc5f52 to your computer and use it in GitHub Desktop.
agent-skills v0.6.0 release announcement: 20 new skill entry points, binary analysis, product and production operations, QA, and release engineering

agent-skills v0.6.0: a bigger, safer toolkit for building and operating AI agents

agent-skills v0.6.0 is out.

This is the release where the collection starts to feel less like a directory of prompts and more like an operating toolkit for serious agent work: reusable methods, executable CLIs, evidence contracts, quality gates, lifecycle handoffs, and production controls.

The release at a glance

  • 20 new skill entry points, including three orchestration bundles
  • 8 existing skill surfaces updated
  • 111 commits, including 98 non-merge commits
  • 64 feature entries, 17 bug fixes, and one revert in the tagged changelog
  • A new binary-analysis system with a deterministic CLI, fake backend, Ghidra adapter, structured JSON contracts, security rules, and auditable reports
  • A much deeper product, production, privacy, migration, resilience, QA, and release-engineering layer
  • CI and agent-readiness work that makes the repository itself more testable and more trustworthy

The collection follows the Agent Skills open format, so the skills are usable by any harness that supports the standard rather than being locked to one vendor.

What is new

1. Binary analysis becomes a real, auditable workflow

The new binary-analysis skill gives an agent a disciplined way to investigate unknown PE, ELF, Mach-O, and firmware files without executing or modifying them.

It ships a non-interactive binary CLI with 37 commands across 14 functional groups, including:

  • Environment diagnosis and bootstrap for Ghidra and PyGhidra
  • Project creation, binary import, metadata, and analysis
  • Sections, entry points, imports, exports, symbols, and strings
  • Function inspection, disassembly, bytes, decompilation, and address maps
  • Cross-references, callers, callees, call graphs, search, and trace workflows
  • Triage, diagnostics, suspicious-API analysis, and capability mapping
  • Markdown, JSON, HTML, and PDF reports with binary SHA-256 and analysis provenance

The architecture separates deterministic observations, rule-derived heuristics, and unresolved questions. It includes a fake backend for offline tests, a Ghidra adapter for real analysis, path and output safety controls, pagination, JSON contract tests, and an optional local worker.

If you need Ghidra-backed analysis, the documented path requires Python 3.12+, Java 21+, Ghidra 12.1+, and PyGhidra 3.1+. The project-management and fake-backend paths remain useful without that stack.

2. Product work gets an evidence-bearing lifecycle

v0.6.0 adds the missing connective tissue between an idea, a roadmap, a shipped capability, and what happened afterward:

  • Product discovery and planning: turn approved requirements into dependency-aware implementation plans with vertical slices, ownership, critical paths, staged rollout, rollback, and verification traceability.
  • Product experimentation: choose methods from interviews and prototypes through concierge tests, fake doors, feature flags, and A/B tests, with guardrails and decision readouts.
  • Adoption and measurement: distinguish acquisition, activation, adoption, retention, and expansion, then connect them to governed event plans, funnels, cohorts, countermetrics, and privacy-aware dashboards.
  • Lifecycle learning: compare expected and observed outcomes, preserve uncertainty instead of laundering it into certainty, and make disciplined continue, improve, harvest, pivot, pause, or retire decisions.
  • Product governance and portfolio work: define decision rights, evidence standards, review cadences, Now/Next/Later roadmaps, strategic-bet criteria, dependencies, and scenario plans.
  • Conditional customer success: adapt success plans, health evidence, renewal signals, escalation, and Voice of Customer to products that actually have recurring human relationships.

3. Production readiness is now a connected system

The new production-focused skills treat launch readiness as a cross-domain decision, not a single checklist:

  • production-readiness defines risk-scaled evidence for Low, Standard, and High releases, with Go, No-go, Defer, and Exception outcomes.
  • production-excellence composes readiness, migration, recovery, capacity and cost, incident learning, QA, security, platform, data, release, and verification evidence into one gate.
  • agent-production-operations gives evaluated agents an operating contract for staged rollout, fallback, cost and latency budgets, tool health, escalation, disablement, and trace-to-eval feedback.
  • migration-engineering covers compatibility windows, dual-running, backfills, reconciliation, cutover, deprecation, and the difference between rollback, roll-forward, restore, and irreversible recovery.
  • resilience-and-recovery turns graceful degradation, restore drills, RTO/RPO decisions, failover exercises, integrity verification, and recovery communication into explicit artifacts.
  • capacity-and-cost-engineering connects demand, peak sizing, degraded-mode capacity, quotas, load testing, unit economics, and SLO-cost trade-offs.
  • incident-learning separates observations, causal hypotheses, and unknowns, then requires owned improvements and evidence of implementation rather than closing the loop with a ticket alone.
  • privacy-engineering makes purpose, lifecycle, access, deletion, isolation, residency, consent, agent traces, and analytics telemetry reviewable engineering concerns.

4. QA becomes more than “run the tests”

qa-methodology was substantially rebuilt with modern traditional QA and SDET references, agentic evaluation design, mutation-guided test hardening, career-level guidance, fillable templates, tested CLI scripts, and output-quality evals.

It now ships 16 references, 6 templates, 3 assets, 2 tested scripts, and 10 evals. The goal is not a green checkbox. The goal is evidence that the test suite detects the failures that matter, that graders are calibrated, and that quality claims remain bounded by what was actually exercised.

The release also adds repository-level linting, formatting, coverage, security configuration, mypy, radon, deptry, broader test isolation, and agent-readiness checks.

5. Release and delivery practice gets its own toolkit

release-engineering brings senior-to-principal release practice into a reusable skill:

  • Release process models and CD pipeline stages
  • Progressive delivery, feature flags, and rollback planning
  • SemVer and Conventional Commit decisions
  • Changelog and release-artifact validation
  • Readiness gates, change governance, compliance evidence, and provenance
  • DORA metrics and release operations
  • Five standard-library Python CLIs, 15 references, 6 templates, 4 assets, and 8 evals

This is the kind of material that helps an agent understand not only how to change a system, but how to decide whether the change is ready to ship.

Every new skill entry point in v0.6.0

Analysis, engineering, and operational practice

Product practice

Orchestration bundles

Install it

The quickest cross-harness path is the open skills CLI:

# Inspect the full catalog, including bundle entry points
npx skills add magnus919/agent-skills --full-depth --list

# Install a standalone skill for Codex
npx skills add magnus919/agent-skills --skill binary-analysis --agent codex --yes

# Install a skill globally for Hermes Agent
npx skills add magnus919/agent-skills --skill release-engineering --agent hermes-agent --global --yes

# Install a bundle entry point
npx skills add magnus919/agent-skills --skill production-excellence --full-depth --agent codex --yes

The collection is also documented for Claude Code, OpenCode, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Hermes Agent, and other Agent Skills-compatible clients.

Read the release

If you are building agents that need to reason about evidence, authority, safety, quality, and the consequences of shipping, v0.6.0 is the place to start.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment