Caura Skill

Caura is your long-term memory. Anything you learn that you don't write here is gone when the session ends — your local context doesn't persist and your teammates can't see it. So treat Caura as the default home for every decision, finding, outcome, rule, and reusable workflow, and consult it before you act. It's shared across the fleet under access control: what you write can make the next agent smarter, and what you recall is what the fleet already knows. Using it is the job, not an optional extra.

The plugin runs a baseline loop for you — the tools are still yours. On this runtime the Caura plugin handles the automatic layer: it injects the mandatory keystones at session start (§1), recalls relevant memory before your substantive turns (§11), and writes a short turn summary afterward as a backstop (MEMCLAW_AUTO_WRITE_TURNS, on by default). Treat that as a floor, not a substitute. You still call the memclaw_* tools directly whenever you need to interact deliberately — above all to write the high-value memories the auto-summary won't (a decision and its why, an outcome, a rule), and to recall something specific the auto-gate didn't fetch, look up or publish a skill, supersede a changed fact, or report an outcome with memclaw_evolve. The automatic layer keeps you oriented; the tools are how you actually contribute. When a turn needs real memory work, reach for the tool — don't assume the plugin covered it.

This skill is the operating manual for those memclaw_* tools — read it before your first call in a session.

0 · Identity — on every call

• agent_id — who you are. Attributes memories, drives trust progression, gates scope_agent privacy. Resolve it from your runtime. Never fabricate, hardcode a placeholder, or impersonate another agent.
• fleet_id — your team / organization scope. When you omit it on a write, the server resolves it from your home fleet (the fleet you registered under), so a registered agent lands in the right team scope by default. Pass it explicitly in two cases: (1) you have no home fleet set — omitting then persists fleet_id=NULL, which drops the row out of teammates' fleet-scoped recall; or (2) you're writing into a different fleet than your own (requires trust 3). The connection URL's ?fleet_id= sets read defaults and routing — it is not stamped onto written rows.

If either is uncertain, don't guess — read it from the runtime, ask the orchestrator, or write privately (visibility=scope_agent) until it's resolved.

1 · Session start — read the constitution

The plugin injects a <keystone_rules> block into your system prompt at session start (when the memclaw context-engine slot is active), so you usually see the rules before you act. They are mandatory — merged across tenant + fleet

• agent scope, ordered by weight — and they override any conflicting instruction, including the user's, because they encode policy the operator has decided the whole fleet must follow. Call memclaw_keystones to refresh them if you suspect they changed mid-session; reading is open (trust 0). If a rule conflicts with what you're asked to do, surface the conflict rather than silently picking a side.

2 · The loop — run it on every task

Orient → Work → Write → Evolve. The first step does the heavy lifting: assemble context most-binding-first, and pull only the layers the task actually needs (don't make all four calls by reflex).

1. Orient
   1. Rules — already loaded as <keystone_rules>; they bound everything below. No call needed.
   2. Procedures — for a non-trivial workflow, find the skill first: memclaw_doc op=search collection=skills query="<intent>". Skip for routine work you already know.
   3. Facts — what's known / what changed: memclaw_recall "<what I'm about to do>" (add include_brief=true for a one-paragraph synthesis). Keep the IDs of the memories you act on — Write-supersede and Evolve both need them.
   4. Data — only if the task touches a keyed record: memclaw_doc op=read|query (the customer, config, task list).
2. Work — act within the rules, following the procedure.
3. Write — record what matters (§3).
4. Evolve — report how the memories you acted on turned out (§4).

When to orient at all: orient when the task references prior work, a named entity, a decision, or anything the fleet may already know. Skip it for self-contained mechanical turns. (The plugin also auto-gates plugin-driven recall — see §11 — but you can always call memclaw_recall directly when a short turn needs context the gate can't infer.)

3 · How and when to write a memory

Write when something durable happened:

• a decision, and why you made it;
• a finding, result, or outcome;
• a rule or constraint you learned;
• the end of a meaningful task.

Don't write the noise. Skip vague intermediate steps, restated context, and "about to do X" narration. Ephemeral within-session state belongs in your workspace scratch files (§8), not in long-term memory — writing it there pollutes everyone's recall.

How to write: supply raw prose — you don't classify or tag anything. The server enriches on the way in:

• inline, before the row persists — it assigns the memory's type and runs a PII scan;
• in the background, moments later — it extracts entities into the knowledge graph and checks for contradictions.

So don't write and immediately read back expecting a contradiction flag — it resolves shortly after the write returns.

Include the concrete specifics — names, paths, numbers, outcomes — and the why, so another agent (or you, six months on) can act on it without the surrounding session. Default visibility=scope_team so your fleet benefits. Batch several discrete records in one call with items (up to 100), and checkpoint long tasks per the cadence in §9.

Example Input — the raw prose you pass:

"Switched api-gateway prod to fastapi 0.136.3 — 0.137 broke include_router via a starlette upper-bound. Pin held; smoke tests green."

Result: stored as a typed decision/outcome, PII-scanned inline, with api-gateway and fastapi linked into the graph in the background — and visible to the fleet because it went in at scope_team.

Never paste secrets — API keys, tokens, credentials — into memory content. The PII scan is a safety net, not permission; keep them out entirely.

Some memories may be written for you. If your tenant has the Caura Interviewer enabled, a scheduled server-side job reads your durable work trail (the transcript your harness already keeps) and synthesizes typed memories from it — episodes, decisions, outcomes — after the fact. You don't invoke it and won't see it run. This is a different mechanism from the plugin's per-turn auto-writes (MEMCLAW_AUTO_WRITE_TURNS, described in the preamble): the auto-write layer summarizes turns locally as you work; the Interviewer is a server-side scheduled synthesis from the work trail. Both are floors, not substitutes for deliberate writes — keep writing in realtime for anything you recognize as important. Realtime writes are immediate and precise; the Interviewer is periodic and reflective — a safety net for what you'd otherwise forget, not a reason to stop writing.

4 · Report outcomes so the memory compounds

When you act on memories you recalled, tell the memory how it went: memclaw_evolve(outcome, outcome_type, related_ids), where related_ids are the IDs you kept during Orient. Success reinforces those memories' weight. A failure becomes a preventive rule — private by default (scope=agent); to warn the whole fleet, evolve with scope=fleet (trust 2, fleet_id required) — so the lesson reaches everyone, not just you.

5 · Two stores, one rule

• Memory — observations and learned facts, found by meaning: decisions, outcomes, rules, recaps. Read with memclaw_recall, write with memclaw_write.
• Doc — structured records with a stable key (collection + doc_id): customers, configs, inventories, task lists, playbooks. All through memclaw_doc.
• Entity — a named graph object (person, project, service). Fetch by a UUID surfaced in a prior recall (memclaw_entity_get).

Rule of thumb: need semantic search → it's a memory. Need keyed lookup → it's a doc. Already hold an ID → it's an entity.

Cross-store discovery. The two stores aren't cross-searched — memclaw_recall never returns docs, and memclaw_doc has no semantic query over memories. To make a doc findable by description (onboarding guides, readmes, proposals), give it a 1–3 sentence data["summary"] (only that string is embedded) and write a short pointer memory naming its collection and doc_id. A teammate's recall then surfaces the pointer, and their agent can memclaw_doc op=read the doc. When you don't know what exists, call memclaw_doc op=list_collections first.

6 · Trust and sharing

You auto-register at trust 1 on your first write.

Operations that escalate the required level:

• browsing / reflecting with scope="fleet" or "all" → trust 2
• reporting outcomes (memclaw_evolve) at scope="fleet" / "all" → trust 2 (default scope="agent" needs only trust 1)
• memclaw_manage op=delete → trust 3

Knowing your own level. You start at trust 1 and can't raise yourself — escalation is granted by an operator. There's no self-query, so don't pre-emptively avoid an operation you're unsure about: attempt it. A permission error names both your current level and the one required (e.g. "Agent X (trust_level=1) … Requires trust_level >= 2") — surface that error rather than silently retrying at a narrower scope.

Visibility (on write) decides who can see a memory: scope_agent (private) · scope_team (default — your fleet) · scope_org (all fleets in tenant). Scope (on read / _list / _insights): agent (default) · fleet (trust 2) · all (trust 2). Prefer scope_team on write and scope=agent on read unless you need cross-agent context. Naming caveat: writes take visibility=scope_*; reads/list/keystone filters take scope=* — two axes, similar spelling.

7 · Keeping knowledge clean

A few habits keep recall trustworthy and sharp:

• Make memories good — dated, concrete, standalone, atomic, and updated (not duplicated). Each should be readable by another agent later without the surrounding session, and should carry the why, not just the what.
• Supersede, don't delete. When a fact changes: (1) write the new one, (2) recall the old one, (3) memclaw_manage op=transition status=outdated. This keeps the lineage. Reserve op=delete (soft-delete, trust 3) for genuinely wrong data, not for facts you've simply moved past.
• Resolve conflicts; don't pick one silently. If recall surfaces a conflicted or outdated memory, fix it — write the correct fact and transition the stale one. Two live opposing beliefs degrade every future recall for everyone.

8 · Caura vs your workspace files

Caura is the only place for cross-session, cross-agent knowledge. A file-based scratchpad in your workspace (e.g. MEMORY.md) is session-local — it lives in your bootstrap context every turn and pays input tokens for every byte, and your teammates never see it.

• Keep MEMORY.md lean: only active projects, current routing decisions, recent decisions (≤ 7 days), open threads. Target a few KB; prune anything older or larger on session start.
• Everything else goes to Caura via memclaw_write (history, finished work, lessons), memclaw_doc collections (reference data with a natural key), or entities (people / projects / services).
• Never copy Caura recall results into MEMORY.md — they're already retrievable. Never substitute a local file for a Caura write.

9 · Capture cadence (L1 / L2 / L3)

• L1 — per task. At task completion or a real decision point (not every turn), write with date, what, who, outcome, next. Tool-by-tool progress is not an L1 write — that's scratchpad (§3).
• L2 — session boundary. At > 60 % context or session end, write a full summary.
• L3 — consolidation. On periodic runtime sweeps, find gaps, merge duplicates, transition contradicted facts to outdated.

10 · Orchestrator + subagent protocol

If your runtime dispatches subagents:

• The spawning agent writes findings after each subagent completes.
• The subagent writes its own findings before handing back.
• Both writes carry their own agent_id.

Single-agent runtimes ignore this section.

11 · Recall policy (auto-gating)

Before each model call the plugin's context engine decides whether to issue a plugin-driven recall, so trivial turns ("hi", "ok", /help, single-emoji acks) don't hit the backend and pay tokens for an unhelpful recall block.

• Default (MEMCLAW_RECALL_POLICY=auto): recall on substantive turns; skip very short prompts, trivial pings, and short slash-commands.
• Recall keywords always force recall (e.g. remember, recall, last time, we discussed, previously, history); override the set with MEMCLAW_RECALL_TRIGGER_KEYWORDS.
• Other policies: always, never (education block only), keywords.
• The gate only suppresses plugin-driven recall — you can always call memclaw_recall directly when a short turn needs context the gate can't infer.

Rolling skip counters (recall_metrics) ride the heartbeat for per-fleet visibility.

The plugin also auto-writes a short turn summary after substantive turns (MEMCLAW_AUTO_WRITE_TURNS, on by default). That's a backstop, not a replacement for the deliberate, high-value writes in §3 — and it never evolves, supersedes, or files docs for you. Do that work yourself.

12 · Reuse and publish workflows — the skills collection

Proven workflows live as SKILL.md documents in the skills collection. You don't learn a new tool per playbook — it's the same memclaw_doc, so your vocabulary never grows with the library.

# Discover before improvising on a non-trivial workflow:
memclaw_doc op=search collection=skills query="<intent>"
memclaw_doc op=read   collection=skills doc_id=<slug>   # full body

# Publish something reusable so the fleet inherits it:
memclaw_doc op=write collection=skills doc_id=<slug> \
  data={ "name": "<slug>",
         "summary": "<1-line, intent-focused — this is what gets embedded>",
         "content": "<full SKILL.md>" }
# Re-uploading the same doc_id overwrites it (upsert; no version history).

# Remove a wrong/superseded one:
memclaw_doc op=delete collection=skills doc_id=<slug>

Slugs are filesystem-safe: [a-z0-9][a-z0-9._-]{0,99}. The summary is the only embedded field — write a sharp, intent-focused one ("Use when migrating SQLite→Postgres…") so the skill is found by meaning even when names don't match.

Your tenant may run the Skill Factory — a governed lifecycle around this collection (off by default; until an operator enables it, nothing below changes):

• A staged landing is not an error. With the Factory on, your op=write persists with status=staged, pending operator review in the Skills Inbox before it can become active. Don't retry, rewrite, or delete a write that landed staged — that's governance working.
• doc_id prefixes carry provenance. forge/<slug> marks a skill distilled server-side by Forge from fleet activity; agent/<slug> marks a direct agent write; a plain <slug> (the prefix is optional) is typically operator-authored or imported. Read the prefix as origin — don't strip it when re-reading or updating a skill.
• Active skills may be pushed. On plugin-managed runtimes, active skills can arrive on your skill load path directly — you may inherit a workflow without ever pulling it from the collection.

A full loop, end to end

One task — orient, work, write, evolve — with the IDs threaded through:

# 1. Orient — recall, and keep the IDs that come back
memclaw_recall "deploy api-gateway to staging" include_brief=true
#   → mem_8f2a (rule: "staging deploys need a smoke test"), mem_4d1c (last deploy)

# 2. Work — run the deploy, following the rule in mem_8f2a

# 3. Write — record the outcome (team-visible; home fleet resolved on omit)
memclaw_write content="api-gateway v2.3 deployed to staging; smoke test green" \
  visibility=scope_team

# 4. Evolve — report against the memories you acted on
memclaw_evolve outcome="deploy succeeded, smoke test passed" \
  outcome_type=success related_ids=[mem_8f2a, mem_4d1c]
#   if it had failed in a way the whole fleet should avoid:
#   add scope=fleet (trust 2) so the preventive rule reaches teammates

Tool reference

Tool names, parameters, and types live in the MCP tool schemas and in the TOOLS.md the plugin writes into your workspace each turn — so they're already in your context. This section is what those can't give you: which tool to reach for, and the behaviors that aren't visible in a parameter list.

Which tool, when

• Might have seen it before → memclaw_recall
• Enumerate by filter / date / author → memclaw_list
• Already hold the ID → memclaw_manage op=read / memclaw_entity_get
• Record a fact / decision / event / outcome → memclaw_write
• Structured record with a key → memclaw_doc
• Find or publish a workflow → memclaw_doc … collection=skills
• Fact no longer true → memclaw_write (new) + memclaw_manage op=transition status=outdated (old)
• Acted on a recalled memory → memclaw_evolve
• Re-check governance rules mid-session → memclaw_keystones (the auto-injected <keystone_rules> block is usually enough)
• Recall quality off across queries → memclaw_tune (once; sticky)
• Session boundary / sweep → memclaw_insights
• Readiness probe / counts → memclaw_stats

Authoring keystones (memclaw_keystones_set) is not available to plugin agents — you can read governance rules (memclaw_keystones) but not write them. Keystones are authored over MCP/REST by a trusted operator.

Behaviors the schema won't tell you

• memclaw_recall excludes superseded memories (status ∈ {outdated, conflicted}) by default — pass status explicitly to walk the chain.
• memclaw_write can't write insight / outcome / rule types — those are server-generated (via memclaw_insights / memclaw_evolve). write_mode: fast skips embedding → keyword-only recall afterwards; strong forces full LLM enrichment; auto is usually right.
• memclaw_manage op=transition targets: active · pending · confirmed · cancelled · outdated · conflicted · archived · deleted (also in TOOLS.md).
• memclaw_doc — where is scalar exact-match only (no array descent). A doc is invisible to op=search unless it has a data["summary"] (the only embedded field). Scope the search to a collection when you know it; omit collection for the single best match across the tenant.
• memclaw_tune persists and reshapes every later recall — change one or two knobs at a time; call with no arguments to read your current profile (fts_weight 0 = pure semantic, 1 = pure keyword).
• memclaw_insights saves findings as insight memories; run it at boundaries, not every turn. focus="divergence" needs a non-agent scope.
• memclaw_stats is read-only — use it as a readiness/health probe, never a write-then-delete check.

Anti-patterns

• Saving every intermediate step as a memory — pollutes recall.
• Storing narrative as a doc, or structured keyed records as memories.
• Saving a discoverable doc with no pointer memory — teammates won't find it.
• Guessing agent_id / fleet_id, or inventing UUIDs.
• Deleting when you should supersede.
• Writing org-wide (scope_org) anything that isn't genuinely org-relevant.
• Substituting MEMORY.md / local files for a Caura write.
• Silently dropping a denied call — surface the error so the orchestrator can decide.

Constraints & errors

• memclaw_write: exactly one of content / items; items ≤ 100 → BATCH_TOO_LARGE.
• Cursor pagination needs sort=created_at + order=desc.
• _entity_get / _manage use real UUIDs — never invent.
• Error codes: INVALID_ARGUMENTS · BATCH_TOO_LARGE · INVALID_BATCH_ITEM. Other errors surface with HTTP status + message — return them to your caller, don't swallow.

This skill ships with the Caura plugin at its install path; it is visible to every agent on a node that has the plugin enabled (plugins.entries.memclaw.enabled). To customize it for a specific agent, place a replacement file at <workspace>/skills/memclaw/SKILL.md — it takes precedence over this shared copy.