From intent-brain
Saves a single fact, decision, pattern, or convention into the governed knowledge brain, or retires outdated memories. Admin-only write side of the brain, invoked via /brain-save.
How this skill is triggered — by the user, by Claude, or both
Slash command
/intent-brain:brain-save [save <fact> | retire <memory-id>][save <fact> | retire <memory-id>]This skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
This is the **write** side of the brain. `/brain` reads; `/brain-save` writes. Use it to tell the brain
This is the write side of the brain. /brain reads; /brain-save writes. Use it to tell the brain
to remember a specific fact going forward — without re-running a full compile — or to retire a memory
that's no longer true.
The brain learns in two ways: the bulk compile ingests a whole corpus at once, and /brain-save
adds (or retires) a single item on demand. Either way, governance stays in code: this skill
proposes a memory to the deterministic curator, which decides what actually gets stored after policy
checks (dedupe, secret-detection). You are saving an item for the brain to keep, not bypassing the
governance pipeline.
disable-model-invocation: true means Claude will not trigger this from conversation — it runs only
when you explicitly type it. Writing to the shared company brain is a deliberate act, not a chat side
effect. The brain API also enforces this server-side: the write tools succeed only for an admin
token, returning 403 otherwise. This skill is the convenient front door; the server is the real gate.
intent-brain plugin is installed with admin role (TEAMKB_ROLE=admin), so the write MCP
tools are registered. A member install never sees them.TEAMKB_API_TOKEN must carry the admin role, or the brain API
rejects the write with 403.In team mode the write tools reach the brain API over the tailnet with the admin's per-user bearer
token (TEAMKB_API_TOKEN), sent as an Authorization: Bearer header. A member token is rejected
server-side with 403. Never hardcode the token; supply it via env or a headersHelper. In local
mode no token is needed.
decision, pattern, convention, architecture, troubleshooting,
onboarding, or reference.teamkb_propose with { title, content, category, filePaths? }. It writes to the spool;
the curator promotes it after policy checks./brain search or teamkb_status).teamkb_transition with { memoryId, to, reason, actor }. Valid moves:
active → {deprecated, superseded, archived}, deprecated → {active, archived},
superseded → archived. Every transition writes a hash-chained audit event.Call teamkb_status to see counts by lifecycle state and recent rejection feedback before or after
a batch of saves.
candidateId and that it is queued for governance review.Save a decision:
/brain-save we're going Apache-2.0 on both flagships so the public can self-host.
→ teamkb_propose({ title: "License: Apache-2.0 on both flagships",
content: "...", category: "decision" })
→ Saved as candidate 4f3a… — queued for governance review.
Retire a superseded memory:
/brain-save retire memory 9c2e… — superseded by the new deploy runbook.
→ teamkb_transition({ memoryId: "9c2e…", to: "archived",
reason: "Superseded by the new deploy runbook", actor: "jeremy" })
→ Memory 9c2e… → archived; audit event written.
| Situation | Response |
|---|---|
Write returns 403 | The token is not an admin token. The gate working as designed — do not route around it. |
| Write tools are absent | The install is member role; only an admin install registers them. |
teamkb_transition rejects the move | The lifecycle state machine forbids it; pick a valid target state. |
| Content may contain a secret | Stop and strip it. Do not rely on the pipeline's secret-detection as the only check. |
reason on a retire must be a real, human-readable justification — it lands in the permanent audit
trail.403 is the system working as designed, not a bug to work around./brain skill (cited, member-safe queries).npx claudepluginhub jeremylongshore/qmd-team-intent-kbCaptures team knowledge at session end by reviewing changes, classifying insights, checking for conflicts, and proposing governed memories via teamkb MCP tools. Use when wrapping up with team-relevant discoveries or importing docs into the brain.
Records user preferences, corrections, and rules into an agent-writable memory layer, and logs whether active preferences were applied or violated when producing durable artifacts.
Stores important decisions, patterns, bug fixes, and lessons into Cortex persistent memory. Use after resolving bugs, making architecture choices, or when the user asks to remember something.