feat(mem-skill): add mem sidecar skill with mode-selection workflow

.agents/skills/mem/SKILL.md

@@ -0,0 +1,119 @@
+---
+name: "mem"
+description: "Use when the user asks for persistent memory, context retention, memory debugging, decision logging, or long-lived project memory. Installs and operates amcbstudio/mem as an external local sidecar (not vendored), then asks the user which memory mode they want before enabling workflows."
+---
+
+# mem Skill
+
+Use `mem` as a deterministic memory sidecar for this repo. It records append-only events and compacts them into readable artifacts for humans and agents.
+
+## Important constraints
+
+- `mem` is local CLI tooling, not a cloud memory API.
+- Do not vendor `mem` into this repository.
+- Install it outside this repo and point to it by absolute path.
+- Treat `mem` as assistive memory; keep operational truth in TinyClaw's structured files/queues/settings.
+
+## First action: ask the user their memory mode
+
+Before installing or writing events, ask one compact question:
+
+"How do you want memory used in this workspace? (A) Dev timeline/audit only, (B) Agent context notes/decisions, (C) Both, (D) Not now)"
+
+Then confirm two policies:
+- retention window (for example: 30 days, 90 days, forever)
+- sensitivity policy (what must never be stored)
+
+If the user does not specify, default to:
+- mode: A (dev timeline/audit)
+- retention: 90 days
+- sensitivity: no secrets, tokens, personal identifiers
+
+## Install (external, from GitHub)
+
+Install under `~/.tinyclaw/tools/mem`:
+
+```bash
+mkdir -p ~/.tinyclaw/tools
+git clone --recurse-submodules https://github.com/amcbstudio/mem.git ~/.tinyclaw/tools/mem
+```
+
+Verify:
+
+```bash
+~/.tinyclaw/tools/mem/bin/mem --help
+```
+
+## Enable in the target repo
+
+From this TinyClaw repo root:
+
+```bash
+~/.tinyclaw/tools/mem/hooks/install.sh
+~/.tinyclaw/tools/mem/bin/mem init
+~/.tinyclaw/tools/mem/bin/mem sync
+~/.tinyclaw/tools/mem/bin/mem show
+```
+
+This creates `.amcb/memory/` artifacts in the repo.
+
+## Usage patterns for TinyClaw
+
+### Mode A: Dev timeline/audit (recommended default)
+
+- Rely on git hooks (`post-commit`, `post-merge`) installed by `mem`.
+- Add occasional notes/decisions:
+
+```bash
+~/.tinyclaw/tools/mem/bin/mem add decision scope="queue routing" rationale="reduce regressions"
+~/.tinyclaw/tools/mem/bin/mem sync
+```
+
+### Mode B: Agent context notes/decisions
+
+- Add explicit events at key moments (incident, routing decision, policy choice).
+- Keep payloads minimal and non-sensitive.
+- Prefer short key/value fields over long prose.
+
+Example:
+
+```bash
+~/.tinyclaw/tools/mem/bin/mem add note agent="coder" topic="memory-strategy" summary="hybrid memory adopted"
+~/.tinyclaw/tools/mem/bin/mem sync
+```
+
+### Mode C: Both
+
+- Combine A + B.
+- Keep `mem` as retrieval/audit context, not as authoritative queue state.
+
+## "Cloud" clarification (be explicit)
+
+If the user asks for cloud memory, clarify that upstream `mem` has no remote runtime service.
+
+Offer two alternatives:
+- Git-backed sharing: commit/push `.amcb/memory/` artifacts to a remote repo.
+- Separate cloud memory service: keep `mem` locally for deterministic logs and integrate a second system for cloud retrieval.
+
+Do not claim native cloud sync for `mem`.
+
+## Operational guardrails
+
+- Never store secrets, API keys, auth tokens, personal identifiers, or private conversation dumps.
+- Keep events compact and typed (`note`, `decision`, `run`, `commit`, `merge`).
+- Run `mem sync` after batched writes.
+- If schema drift appears, keep going but add a follow-up `note` documenting the change.
+
+## Quick checks
+
+```bash
+test -d .amcb/memory && echo "mem initialized"
+jq -r '.counters.events_total' .amcb/memory/state.json
+tail -n 10 .amcb/memory/events.jsonl
+```
+
+## When not to use
+
+- Do not use `mem` as TinyClaw's queue, scheduler, or pairing source of truth.
+- Do not use `mem` as a replacement for repo docs like `AGENTS.md` and `SOUL.md`.
+- Do not use `mem` if the user forbids local event storage.