Relic Team Memory
Turn approved team traces into source-linked operational memory. Do not summarize everything: extract decisions, context, owners, unresolved questions, and reusable lessons.
Safety gate
Before reading material, confirm data-owner permission, participating channels, date range, audience, retention period, redaction rules, and a withdrawal route. Never ingest private conversations, credentials, customer secrets, or an entire workspace by default.
Core outputs
- decision log with alternatives, rationale, source ids, owner, and status;
- action register with accountable person, due date, and evidence;
- unresolved questions and assumptions;
- project retrospective with what changed and why;
- onboarding packet for a new teammate;
- manifest describing source scope and review date.
Operating loop
- inventory sources and classify sensitivity;
- normalize dates, speakers, channels, and document links;
- separate fact, decision, opinion, action, and open question;
- preserve disagreement and record conditions;
- ask the team owner to review before publication; unresolved items must enter a follow-up queue;
- remove or update entries when a source is withdrawn or superseded.
Every decision entry must contain id, title, decision, status, owners, evidence, and review.reviewer. Allowed states are proposed, accepted, superseded, unknown, and needs-verification. unknown and needs-verification entries must include a concrete follow_up_question and cannot be presented as settled team policy.
Image placeholder
TODO(image): add an authorized decision-flow diagram after the product visual system is approved.