By Skobyn
A 'Memory and a Mirror' agentic harness for long-running unattended coding sessions: a prunable structured learnings file read at session start, a proof-discipline gate that blocks 'done' claims without fresh cited evidence, and a fresh-context reviewer subagent that returns only a gap list against spec. Chains into one plan -> act -> prove -> review -> remember loop via /mirror-loop. Model-agnostic by construction.
Use at the start of every coding session to read the project's LEARNINGS.md in full before taking any action, and at the end of every session to update it — applying relevant lessons, adding at most one new block for a genuinely new durable lesson, and actively pruning or merging stale/contradictory blocks so the file never becomes append-only noise. Trigger phrases: "start a session", "begin the task", "remember step", "update learnings", "session end".
Use to run or resume an unattended coding session as a plan -> act -> prove -> review -> remember cycle, chaining the memory-file, proof-discipline, and fresh-context-reviewer disciplines into one durable loop with a deterministic halt condition. Also use immediately after any context-compaction event to resume correctly from on-disk state alone. Trigger phrases: "run the mirror loop", "start an unattended session", "resume after compaction", "chain plan act prove review remember".
Use immediately before making, or considering making, any claim that a task, subtask, or fix is "done", "complete", "fixed", or "passing". Blocks the claim unless it is backed by command or test output that was produced in direct response to this specific claim — not narration, not code inspection alone, and not evidence from an earlier turn that predates the most recent relevant change. Trigger phrases: "report completion", "mark done", "is this fixed", "declare complete".
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
npx claudepluginhub skobyn/upskill-me --plugin memory-mirror-harnessUse when a tech lead, senior developer, or author needs a structured prep briefing before a live synchronous peer code review — given a diff or codebase path, produces intent narrative, correctness questions, design-fit observations, advisory security/standards/performance findings, a verdict suggestion, and a re-review action-item checklist. Read-only; never writes code or mutates anything.
Traxxall author-side delivery skill set. Pull/analyze ADO work items into structured intake briefs (flagging Requires-Specification when under-specified), author implementation specs with blind-QA-ready acceptance criteria, draft/create PBIs and child dev Tasks with compliant users/{short}/{WI#}-… branches, and outline UnitTests.Core/TransactionScope dev tests — all under Traxxall conventions (branch/PR rules incl. VS403660, PBI/Bug/Task model, area Traxxall Solution, AI tags, dual-framework legacy vs modern, egress-safe dev.azure.com auth). Drafts by default; creates ADO items only on explicit go. Never authors QA tests (separate traxxall-delivery-qa plugin) and never runs destructive test projects.
Traxxall blind, adversarial QA skill set — deliberately isolated from implementation. From the acceptance criteria + the running TEST app ONLY (never the diff, PR, product source, or dev-side tests), authors adversarial Playwright suites covering happy/edge/negative/abuse/cross-module cases with role/text locators, deterministic waits (no hard sleeps), storageState auth, and trace/screenshot/video on failure; records PASS/FAIL/ERROR for the PH QA / Ready-to-Test flow. Refuses any leaked implementation context (isolation gate). Targets TEST only, never prod.
Reviews a target repository and produces two grounded board-review deliverables: a self-contained JSSI-branded interactive HTML viewer (16 diagram tabs, built by editing a copy of the approved reference template verbatim) and docs/architecture/index.md (narrative plus two strictly separate tables — Findings for code-level gaps, Open Decisions for architecture choices needing a human ruling). Supports a silent/autonomous generation mode and a distinct post-complete interview mode that walks a reviewer through Open Decisions one row at a time.
71-skill bug-hunting & external red-team bundle for Claude Code — 48 hunt-* web/vuln-class + framework skills, enterprise platform attack chains (M365/Entra, Okta, SharePoint, vCenter, SSL-VPN, APK), recon/OSINT, reporting & validation gates, and Burp MCP integration. Skills auto-load by topic; 15 slash commands included.
Consult multiple AI coding agents (Gemini, OpenAI, Grok, Perplexity, plus codex, antigravity, and grok CLIs when installed) to get diverse perspectives on coding problems
Comprehensive skill pack with 66 specialized skills for full-stack developers: 12 language experts (Python, TypeScript, Go, Rust, C++, Swift, Kotlin, C#, PHP, Java, SQL, JavaScript), 10 backend frameworks, 6 frontend/mobile, plus infrastructure, DevOps, security, and testing. Features progressive disclosure architecture for 50% faster loading.
Comprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality review
Access thousands of AI prompts and skills directly in your AI coding assistant. Search prompts, discover skills, save your own, and improve prompts with AI.
Upstash Context7 MCP server for up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your LLM context.
Complete developer toolkit for Claude Code