Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/opita-code/dark-memory-mcp/dark-memorynpx skills add Opita-Code/dark-memory-mcp --skill dark-memorygit clone --depth 1 https://github.com/Opita-Code/dark-memory-mcpWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00353 | $0.07831 |
| Opus 5 | $0.00177 | $0.03915 |
| Sonnet 5 | $0.00071 | $0.01566 |
| Haiku 4.5 | $0.00035 | $0.00783 |
Grade A, and why
dark-memory scanned grade A with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
not a code bug. Verify with curl before touching the Go code. How it starts
The opening of the file, as written. The whole thing — 469 lines — stays where its author put it; the contents beside it link to each section on GitHub.
dark-memory-mcp — Harness Skill
Targets dark-memory-mcp v2.13.0 — 52 canonical + 3 red-team tools, schema v25.
Self-update: call
dark_memory_health_ping. Ifserver.version > target_version, the server was upgraded. Re-ingest viadark_memory_agent_bootstrap(surface='system_prompt')and check the latest GitHub release for an updated copy of this skill. Ifserver.version < target_version, your binary is stale — rebuild from the repo tag that matches this skill.
TL;DR — 8 rules that prevent 90% of friction
| # | DO | DON'T |
|---|---|---|
| 1 | session_start before any write tool |
Call write tools without a session (ErrFrameStaleTooFar) |
| 2 | health_ping first — cheap, no side-effects |
Assume the MCP is healthy after a restart |
| 3 | agent_memory_recall or research_recall first, then webfetch last |
Jump to web search for facts you already stored |
| 4 | agent_memory_save for persistent knowledge (survives sessions). vibe_publish for work-in-progress (spec + judge + drift gate) |
Mix them up: specs don't survive, memories don't get judged |
| 5 | consensus(n=3) on high-stakes claims, judge for routine |
Ship debatable claims without a second opinion |
| 6 | On drift_detected: fix and re-publish. On needs_human: STOP |
Ignore drift verdicts — they block the vibe loop |
| 7 | session_close(reason=clean) at the end |
Leave sessions dangling (sweeper closes them, but loses context) |
| 8 | Sub-agent spawn: delegate_intent for routing decisions, mindset_apply for prompt composition, agent_memory_delegate for context handoff |
Spawn sub-agents blind — they inherit no dark-memory context and their writes pollute your ContextRecap |
0. Decision tree — which tool when
| Need | Tool |
|---|---|
| Start tracking work | session_start → get session_id back |
| Store a fact, decision, finding, or todo for later | agent_memory_save(kind=...) |
| "What do we know about X?" | agent_memory_recall(query="X") (FTS5-BM25 ranked) |
| Browse all memories by type | agent_memory_list(scope=project, kind=...) |
| Research something new (CVE, paper, domain, IP) | research_topic(query=...), then research_recall to retrieve |
| Create a governed spec with tasks | vibe_spec(vibe_case=C1..C7, tasks=[...]) |
| Publish an artifact under a spec + get drift verdict | vibe_publish(artifact={...}, spec={vibe_case:...}) |
| Self-check a claim | `judge(eval_type=drift_judge |
| N-shot verification | consensus(eval_type=..., n=3) |
| "Should I delegate this work to a sub-agent?" — decide + plan + prompt | delegate_intent(vibe_case=C7, task_description="...") |
| "Just give me a sub-agent prompt for X" — compose only | mindset_apply(vibe_case=C1..C7, task_description="...") |
| Hand off dark-memory context to a sub-agent | agent_memory_delegate → inject delegation_context into sub-agent prompt |
| Isolate sub-agent writes (C2) | subagent_register before spawn, subagent_unregister after |
| Diagnose "the MCP seems broken" | health_ping → error_summary → error_list |
| Check drift status of a published artifact | pipeline_status(artifact_id=N) |
| Accept/reject a drift report | `resolve_drift(drift_id=N, decision=accept |
| See what changed recently | writes(session_id=...) or recall(scope=session) |
| Check schema version / run migrations | admin_schema_status → admin_migrate |
| Create a new tenant/project | project_create(project_id="...", display_name="...") |
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 469 lines · 353 tokens per session scan A 1a238c2ca9c9
dark-memory is a skill published in the GitHub repository Opita-Code/dark-memory-mcp (0 stars, last pushed 12d ago), licensed MIT. It adds 353 tokens to every session and 7,831 once invoked, about $0.0018 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
agentbook-version-control
Coordinate Git and Cross-Session Memory as version control for agent work. Use when Codex needs AgentBook status, checkpoints, history, resume or handoff, provenance, isolated agent approaches, semantic comparison, or an evidence-backed game-development tuning and playtest iteration with an explicit keep or revert…
csm-continuity
Use the complete native Cross-Session Memory (CSM) runtime for continuity, memory governance, living state, beliefs, self-model, AgentBook, checkpoints, context cache, goals, work ledger, compaction, re-entry, or handoff across Codex tasks.
csm-governance
Safely deduplicate, review, merge, and archive Cross-Session Memory. Use when memory needs cleanup, dedup, merge, archive-candidate review, or recall-quality assessment.
csm-handoff
Checkpoint expensive state and write a concrete handoff before stopping, compaction, or transferring work in Claude Code.
webmcp-nexus
当接入 webmcp-nexus-sdk 的项目中需要新增 WebMCP 工具函数、将现有业务方法改造为 WebMCP 工具函数，或配置 vite-plugin-webmcp-nexus / webpack-plugin-webmcp-nexus 时使用本技能。本技能约束工具函数的签名、JSDoc、TS 类型写法，并提供零风险的改造流程（仅改签名与注释，不动原方法业务逻辑）。触发示例："新增一个 WebMCP 工具"、"把这个函数改造成能被 AI Agent 调用的工具"、"让 xxx 函数接入 WebMCP"、"配置 webmcp 构建插件"。.
admin-data-table
Use when creating or editing an admin list/table page in web/src, wiring any paged query, or implementing a paginated repository/endpoint in api/. Covers hook choice (usePagedQuery vs useClientPagination), PageQuery type conventions, the backend paging contract, and the DataTable min-h-0 scroll chain.