菜单
Inspect unified-memory retrieval analytics — summary, recent runs, method rollups, current-vs-grep-sidecar comparison, health, and markdown/JSON export from the live Atlas memory_ops store. Read-only and redacted.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Groom rough software ideas into source-backed implementation specs with original intent, capability discovery, tracker hierarchy, QA, Gherkin, and Notion documentation projection.
Append environment variables to ~/.zshenv on bigmac, genomesbox, and the Mac laptop in one step. Use when the user pastes `KEY=value` pairs and asks to add them to zshenv, or when setting secrets that need to be available on all three hosts. ssh aliases `bigmac`, `genomesbox`, and `mac` are already configured where needed.
Commit all local changes in logical groups until the repository is clean. Use when the user asks to commit everything, make all needed commits, or cleanly commit the full worktree without pushing.
Help users operate the published Composio CLI to find the right tool, connect accounts, inspect schemas, execute tools, subscribe to trigger events with `composio listen`, script workflows with `composio run`, and call authenticated app APIs with `composio proxy`. Use when the user asks how to do something with `composio`, wants to run a known tool slug, needs to discover a slug with `composio search`, fix a missing connection with `composio link`, inspect tool inputs with `--get-schema` or `--dry-run`, troubleshoot top-level CLI flows, or explicitly needs `composio dev` guidance.
Use when the user wants to work a LOS Jira ticket or non-trivial LOS task in a dedicated fast worktree, especially when they mention /workon, /orchestrate, fast worktree, use a worktree, or parallel agents.
Use when working or resuming LOS Jira implementation in a fast worktree with minimal chat output, especially after noisy /workon loops, repeated stack setup, test polling, or context/token burn.
| name | memory-analytics-viewer |
| description | Inspect unified-memory retrieval analytics — summary, recent runs, method rollups, current-vs-grep-sidecar comparison, health, and markdown/JSON export from the live Atlas memory_ops store. Read-only and redacted. |
Use when the user asks /memory-analytics, wants to see how the unified memory
router is retrieving (grep vs typed vs vector), compare the current route to the
grep sidecar, check retrieval-analytics freshness, or export a tuning report.
Live retrieval analytics are in Atlas, database losmon, collection
memory_ops — op: "read" rows with provenance.retrieval. genomesbox is the
host that can reach Atlas. The viewer is read-only and redacted: it never writes
the database and never emits the connection string.
harness/commands/os-memory-analytics.md for the command contract.ssh genomesbox true).Pick the subcommand from the user's intent: summary, recent, methods,
compare, health, or export.
Run the wrapper, which executes the report script on genomesbox:
harness/bin/agentic-os-memory-analytics summary --project losmon --since 7d
or directly:
ssh genomesbox 'cd /home/genome/projects/losmon && \
pnpm -s tsx scripts/memory/retrieval-analytics-report.ts <cmd> --since 7d'
For export, write artifacts to the active work item's
artifacts/retrieval-analytics/ folder (the wrapper copies them back from
genomesbox).
Always surface the data-target class (Atlas vs local/dev) and newest-row freshness so live and dev data are never confused.
When summarizing the grep-vs-current comparison, distinguish hit COUNTS from
answer QUALITY — more grep hits is not evidence to promote grep past
compare. Defer promotion to a labelled benchmark
(scripts/memory/retrieval-bench.ts).
GET /api/memory-view/retrieval/{summary,runs,methods,experiments,export}.scripts/memory/retrieval-bench.ts.scripts/memory/memory-health.ts.