| name | flux-remember |
| description | Smart "remember" routing — when user says "remember X", "don't forget X", or "keep in mind X", this skill decides whether to store in CLAUDE.md (short actionable rules for every session) or the brain vault (deeper context, decisions, business knowledge). Routes through flux-brain's "Remember" flow. Triggers: /flux:remember, "remember", "don't forget", "keep in mind", "note that". |
| user-invocable | true |
Remember
Routes "remember X" requests to the right destination: CLAUDE.md or .flux/brain/.
This is a thin routing skill that delegates to the flux-brain skill's "Remember" flow after classifying the input.
Detection
Trigger on any of these patterns in user input:
- "remember ..."
- "don't forget ..."
- "keep in mind ..."
- "note that ..."
- "from now on ..."
- "always ..." (when giving a rule, not asking a question)
- "never ..." (when giving a constraint, not asking a question)
Do NOT trigger if the user is asking a question ("do you remember...?", "can you remember...?").
Session Phase Tracking
On entry, set the session phase:
PLUGIN_ROOT="${DROID_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT:-$(git rev-parse --show-toplevel 2>/dev/null || pwd)}}"
[ ! -d "$PLUGIN_ROOT/scripts" ] && PLUGIN_ROOT=$(ls -td ~/.claude/plugins/cache/nairon-flux/flux/*/ 2>/dev/null | head -1)
FLUXCTL="${PLUGIN_ROOT}/scripts/fluxctl"
$FLUXCTL session-phase set remember
On completion, reset:
$FLUXCTL session-phase set idle
Process
Step 1: Extract the thing to remember
Parse the user's input to extract the core content. Strip the trigger phrase:
- "remember that we use pnpm" → "we use pnpm"
- "don't forget: the API key is in Vault" → "the API key is in Vault"
- "always run tests before committing" → "run tests before committing"
Step 2: Classify destination
Use this heuristic to pre-select CLAUDE.md vs .flux/brain/:
→ CLAUDE.md (short, actionable, every-session rules):
- Commands and how to run them ("use
pnpm test", "run make build")
- Hard constraints ("never import from
legacy/", "always use TypeScript")
- Code style rules ("use camelCase for APIs", "4-space indent")
- Build/test/deploy instructions
- Things that prevent common mistakes on ANY task
- Patterns: "always X", "never Y", "use X instead of Y", "run X before Y"
→ .flux/brain/ (deeper knowledge, context, decisions):
- Why a decision was made ("we chose Supabase because...")
- Business context ("Sarah is the PM", "we have 500 users", "launch is March 15")
- Team information ("John handles frontend", "Maria approves API changes")
- Domain knowledge ("a 'workspace' is a tenant in our system")
- Pitfalls and lessons learned ("migration script doesn't handle NULLs")
- Architecture decisions ("we use event sourcing for audit trail because...")
- Anything that needs explanation beyond a single bullet point
Rule of thumb: If it fits in one bullet point and the agent needs it on every task → CLAUDE.md. If it needs context, rationale, or only matters in specific situations → .flux/brain/.
Step 3: Delegate to flux-brain
Invoke the flux-brain skill's "Remember" flow with:
- The extracted content
- The pre-selected destination (CLAUDE.md or
.flux/brain/)
- The user's original phrasing (for context)
The flux-brain skill handles:
- Asking the user to confirm the destination (with AskUserQuestion)
- Writing to CLAUDE.md (finding the right section, appending as bullet)
- Writing to
.flux/brain/ (asking category, creating file, updating index)
Gotchas
- Not every "remember" belongs in
CLAUDE.md. Short every-task rules go there; contextual or explanatory knowledge belongs in .flux/brain/.
- The user's phrasing is a clue, not the final storage schema. Classify first, then delegate to
flux-brain for the actual durable write.
- Avoid storing one-off session state as memory. If it will not matter in future sessions, do not persist it.
- When writing to CLAUDE.md, task-specific rules should go inside existing
<important if> conditional blocks when a matching one exists. See flux-brain Step 3a for details.
Examples
| User says | Destination | Reasoning |
|---|
| "remember to always use pnpm, not npm" | CLAUDE.md | Short rule, every task |
| "remember that Sarah is the PM and approves all UX changes" | .flux/brain/business/ | Team context, specific situations |
| "don't forget: the deploy script needs AWS_REGION set" | CLAUDE.md | Command/constraint, prevents mistakes |
| "keep in mind we chose Postgres over Mongo because of ACID compliance" | .flux/brain/decisions/ | Decision with rationale |
| "remember our API rate limit is 100 req/min per user" | .flux/brain/codebase/ | Technical context, not every-task rule |
"always run pnpm lint before committing" | CLAUDE.md | Command, every task |
| "note that the billing team uses 'subscription' not 'plan'" | .flux/brain/business/glossary | Domain language |
