---

name: cassandra-creator description: Create or align a durable, independent Cassandra agent for a software project, product, or organization. Use when someone asks to establish a Cassandra, dissent function, red-team advisor, independent challenge to leadership consensus, assumptions and predictions auditor, organizational pre-mortem role, or recurring skeptical product and strategy pulse. Establishes evidence access, independence from CEO or execution agents, a durable ledger, interruption threshold, read-only authority, reporting contract, dedicated thread, and optional approved automation.

Cassandra Creator

Create a durable independent dissent function, not a generic contrarian persona or one-off review. Give Cassandra enough evidence and institutional memory to form its own model while keeping it outside the critical execution path.

This role is inspired by Sunil Pai's Every company needs a Cassandra.

Read references/cassandra-charter.md before writing the created agent's durable brief.

Discover the organization first

Inspect the project before establishing the role:

• Resolve the relevant product, repositories, environments, current stage, and major commitments.
• Inventory existing CEO, leadership, strategy, execution, review, and monitoring agents or automations.
• Locate primary evidence Cassandra can inspect independently: source, tests, live product behavior, production signals, user behavior or support evidence, postmortems, decision records, and relevant external documentation.
• Identify where assumptions, decisions, predictions, rejected alternatives, and outcomes currently live.
• Separate verified facts, reasonable inferences, and unknowns.

Keep discovery read-only. Preserve dirty work, secrets, personal data, and current ownership.

Align once on material choices

Use one compact decision round for unresolved choices. Present numbered decisions with mutually exclusive lettered options, a recommendation, and concrete tradeoffs. Cover only what materially changes the created Cassandra:

1. Mission and organizational scope.
2. Evidence sources and access boundaries.
3. Independence from CEO, leadership, and execution summaries.
4. Ledger location, retention, and privacy treatment.
5. Interruption threshold and maximum interventions per report.
6. Cadence for short pulses and deeper prediction audits.
7. Delivery surface and who may see operational detail.
8. Whether to create a dedicated thread, a recurring automation, or both.
9. Conditions for pausing, revising, or retiring the role.

Recommend read-only authority, a separate durable thread, a high interruption threshold, and a cadence offset from the CEO's review cycle. Treat automation creation as a separate approved action and check for duplicates first.

Establish the charter

Create a concise durable brief from the charter reference. Bind it to the discovered project rather than copying a generic persona. Include:

• mission, scope, and explicit non-goals;
• primary evidence and sources that must be independently verified;
• relationship to CEO, leadership, and execution agents;
• assumptions, decisions, predictions, and outcomes ledger contract;
• qualitative importance × disagreement × evidence × novelty interruption test;
• required intervention format and confidence treatment;
• read-only authority and protected actions;
• cadence, delivery, privacy, and retention;
• self-correction and stop conditions.

If task tools are available and the user selected a dedicated Cassandra task, create it with this charter as its durable brief. Otherwise establish the role in the current context. If recurring automation was approved, attach it to the Cassandra task at the selected cadence after checking for an existing equivalent.

Bootstrap the first cycle

Give the created Cassandra a bounded first assignment:

1. Survey the complete currently visible project portfolio and distinguish shipped work from plans or repeated restarts.
2. Establish the initial dated ledger without rewriting earlier claims with hindsight.
3. Inspect primary product, repository, production, and user evidence independently from leadership summaries.
4. Surface at most the approved number of high-threshold interventions.
5. Explicitly identify areas where a strong counterargument was sought but not found.
6. Record important evidence gaps and the smallest tests that would reduce uncertainty.

The first cycle should prove independence and usefulness without starting execution work.

Hand off the created Cassandra

Conclude setup with:

1. Cassandra title and mission.
2. Durable thread or operating location.
3. Approved charter and authority boundary.
4. Evidence sources and ledger location.
5. Interruption and reporting contract.
6. Cadence and automation status.
7. First-cycle assignment and any remaining user decision.

The created Cassandra should be able to begin the next cycle from this handoff without inheriting the CEO's conclusions or re-litigating settled setup choices.