jaroslavsoucek-art

jaroslavsoucek-art/Giovanni

Agent Claude Code

Custom subagents for Giovanni. Each lives at .claude/agents/ .md and is spawned via the Agent tool (or via slash commands that route through prediction-runtime).

jaroslavsoucek-art/Giovanni

Agent Claude Code

Specialist architect that extracts adversarial review patterns from source AI Chief of Staff implementation — workflow, trigger patterns, SHIP/REWRITE/KILL verdict format, default-critical-mode policy, generic adversarial-reviewer agent definition. Reads source workflow + agent, writes generic adversarial workflow doc…

jaroslavsoucek-art/Giovanni

Agent Claude Code

Adversarial / red-team review of a draft (deliverable, decision record, position paper, fundraise narrative, board material, public communication, predictive forecast, constitution patch). Default mode is critical, not advisory. Trigger on [REVIEW] tags in draft, "review this", "redline", "before sending", "tear this…

jaroslavsoucek-art/Giovanni

Agent Claude Code

Run semantic consistency checks across memory, constitution, agent roster, decision records, and the decided-terms registry. Surfaces contradictions and drift that deterministic lint can't catch (e.g. memory blocker contradicts constitution claim; agent roster description doesn't match agent file capabilities; a…

jaroslavsoucek-art/Giovanni

Agent Claude Code

Pre-share QA pass on a deliverable (file in deliverables/ or proposed external message). Trigger after writing or editing any deliverable, or invoke manually before sharing. Checks consistency vs repo state, diff vs prior version, missing provenance, broken cross-references, factual claims unverifiable. Returns SHIP /…

jaroslavsoucek-art/Giovanni

Agent Claude Code

Specialist architect that extracts daily-digest workflow patterns from source AI Chief of Staff implementation — 12-step procedure, multi-source parallel pull orchestration (via source-puller fan-out), drift detection with ack flow + 7d expiry, briefs auto-gen for ≤48h events, predictive integration (shadow generation…

jaroslavsoucek-art/Giovanni

Agent Claude Code

Specialist architect that extracts governance-layer patterns from a source AI Chief of Staff implementation — living constitution pattern, INDEX/MAP auto-regen scripts, audit cadence hooks, hard-limit enforcement, classification rules, authority hierarchy. Reads from read-only source snapshot, writes to Giovanni's…

jaroslavsoucek-art/Giovanni

Agent Claude Code

Proactive external competitive / market intelligence scan. Default mode = periodic sweep of competitor space (markets × layers per scope config). Focused mode = topic / market / competitor deep-dive on demand. Returns structured memo with material shifts + framework implications + action verdict. Strictly external…

jaroslavsoucek-art/Giovanni

Agent Claude Code

Specialist architect that extracts memory-layer patterns from a source AI Chief of Staff implementation and produces sanitized generic templates + schema documentation. Reads from a read-only source snapshot, writes to Giovanni's memory/ subtree. Critical mode default — flags over-engineering and domain-bleed.

jaroslavsoucek-art/Giovanni

Agent Claude Code

Specialist architect that extracts predictive-layer patterns from a source AI Chief of Staff implementation — branch-out simulation (3 tiers, max t+2 horizon, hard stop on shallow actors), shadow hypotheses (invisible at generation, quarterly review, anti-self-fulfilling), calibration scoring (per-actor monthly…

jaroslavsoucek-art/Giovanni

Agent Claude Code

Executes the predictive-layer slash commands in isolated context — /branch-out (active simulation), /shadow-review (quarterly verdict pass with adversarial lookback), /calibration-report (monthly accuracy aggregation). Carries prediction-architect's 8 binding principles verbatim. Sub-mode selection driven by caller.…

jaroslavsoucek-art/Giovanni

Agent Claude Code

Bootstrap a new stakeholder profile OR refresh an existing one in memory/stakeholders/ .md. Trigger phrases — "bootstrap stakeholder X", "refresh profile X", "who is X" (when asked about a person not yet profiled). Pulls cross-source signals (chat / email / calendar / project tracker / version control), fills the…

jaroslavsoucek-art/Giovanni

Agent Claude Code

Investigate an external topic via web search + fetch, then return a structured memory note + optional constitution patch proposal. Use when the question requires sources outside the repo (regulation, market data, vendor capability, competitor moves, pricing). Trigger phrases — "investigate X", "research X", "what is…

jaroslavsoucek-art/Giovanni

Agent Claude Code

Specialist architect that ships final slash command implementations wiring templates + worker agents + workflows together. Owns /digest, /branch-out, /shadow-review, /calibration-report, /consistency-check, /market-radar, /review, /redline. Defines argument parsing conventions and command registry. Final specialist …

jaroslavsoucek-art/Giovanni

Agent Claude Code

Pull a SINGLE source for a specified time window and return structured findings. Use for parallel fan-out in digest workflows. Caller specifies sourcetype + identifier + window. Returns plain markdown bullets — no synthesis, no prioritization, no recommendation. Generic over sourcetype enum (chat-platform / email /…

jaroslavsoucek-art/Giovanni

Agent Claude Code

Specialist architect that extracts per-stakeholder profile schema patterns from a source AI Chief of Staff implementation — identity, role, sentiment trajectory, communication style, active threads, predicted reactions, watch points — and produces sanitized generic templates + workflow documentation. Reads from…

jaroslavsoucek-art/Giovanni

Agent Claude Code

Specialist architect that produces the generic subagent roster — source-puller, researcher, profile-bootstrap, deliverable-reviewer, consistency-checker, market-radar, prediction-runtime — based on patterns extracted from source AI Chief of Staff implementation. Plus agent design patterns documentation (when to spawn…

jaroslavsoucek-art/Giovanni

Command Claude Code

Giovanni's invokable surface. Each command is a thin shell that runs pre-flight, parses args, then routes to an agent (or to a workflow procedure). The mechanics live in the agents and workflows; these files are the invocation contracts.

jaroslavsoucek-art/Giovanni

Command Claude Code

Active predictive simulation for a specific situation — routes to prediction-runtime in branch-out mode. Hard-stops on shallow actors.

jaroslavsoucek-art/Giovanni

Command Claude Code

Monthly calibration report — routes to prediction-runtime in calibration-report mode. Aggregates per-actor + per-tier accuracy, updates actor-scores.yaml.

jaroslavsoucek-art/Giovanni

Command Claude Code

Run semantic consistency checks across memory, constitution, agent roster, decision records, and the decided-terms registry. Routes to consistency-checker agent. Read-only — proposes diffs, never applies them.

jaroslavsoucek-art/Giovanni

Command Claude Code

Manual per-finding triage of a /consistency-check report — accept-diff / reject / defer / false-positive verdicts, precision tally update. Interactive main-thread workflow, no agent spawn.

jaroslavsoucek-art/Giovanni

Command Claude Code

Run the daily digest — parallel source pull, synthesis, drift detection, brief generation, shadow lookback, render to chat.

jaroslavsoucek-art/Giovanni

Command Claude Code

Proactive external competitive / market intelligence scan. Routes to market-radar agent. Default = periodic sweep; focused = topic / market / competitor deep-dive.