antreasantoniou 发布的技能

Map repositories into focused context packs

Design and compose multi-agent graphs for correctness, coverage, or creativity. Use when a task benefits from isolated proposals, explicit arbitration, adversarial verification, committees, recursive review, cross-modal checks, or saturation loops. Applies across agent runtimes; the included JavaScript workflow is one adapter.

Reconstruct missing project history safely

One secure Bitwarden unlock per 24 hours

Challenge ML claims with code and falsification

Design or audit self-describing ML run metadata attached at initialization, including intent, code and data identity, resume lineage, and evidence limits. Use when wiring a tracker, introducing a run naming pattern, or investigating why a run exists.

Turn conversations into grounded decisions

Govern and actively monitor collaborative agent work

Durable memory for projects and research

Run a test-sandwiched multi-agent build pipeline

Narrative-aware, approval-gated inbox triage

Preserve durable context across every Archivum

Propose a canonical home, concrete consumers, and any justified runtime home for a reusable tool, library, skill, document, or workflow. Use after building a reusable capability or when asked where it should live; execution requires approval of exact targets and actions.

Use GCP without recurring browser OAuth

Find relevant people and grow your professional network

Sync Claude and Codex project progress

Run a Byzantine 2+1 plan before consequential work and a fresh, plan-blind reality check afterward. Use when the user asks for Plus Ultra, maximum rigor, independent proposals, enforced verification, or help responding to a Plus Ultra hook gate. Claude Code can enforce the loop through the included hook adapter; on Codex and other hosts it is a convention unless separately wired and verified.

Evidence-backed browser and API test plans

Review author-written applications, essays, bios, proposals, and important correspondence through explicitly synthetic, contrasting review lenses while preserving the author's voice and final editorial authority. Use for substantive critique and author-led revision, not impersonation or automatic submission.

Prepare and verify a checkpoint-based handoff of agent work to another host or session while preserving context, permissions and single-writer ownership. Use when the user requests moving ongoing work or continuing after a local disconnect; this is not live process migration.

Design and visually review print-ready PDFs

Build your digital-twin writing voice

Design and configure self-contained recurring agent checks with explicit targets, authority, change-sensitive reporting and stop conditions. Use for requested monitoring, reminders or follow-ups; use the included Meta-Pulse guidance when reviewing existing watches.

Find and open locally running web interfaces

Explore a design or strategy decision through distinct straightforward, hybrid and creative approaches, with concrete tradeoffs and a reasoned recommendation. Use when the user wants options before choosing, not after they have already decided.

Track per-project Claude usage estimates and local GPU budgets, preview admission, atomically reserve capacity, and reconcile actual usage with a local dashboard.

Maintain an explicit Markdown commitments ledger with a local cockpit for NOW, next actions, deadlines, waiting items, workstream states and inbox capture.

Validate, plan and explicitly dispatch provenance-linked ML experiments using the bundled Python runtime, with budget reservations and durable result bundles.

Back up selected local Claude and Codex session stores additively into an explicitly chosen private destination, with a no-write preview.

Build a source-indexed navigation layer over a private Claude JSONL session, preserving evidence locations and uncertainty without reproducing private reasoning.