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 agents/sheawinkler/contextlattice/universalgit clone --depth 1 https://github.com/sheawinkler/ContextLatticeWhat 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 | $0.00000 | $0.01380 |
| Opus 5 | $0.00000 | $0.00690 |
| Sonnet 5 | $0.00000 | $0.00276 |
| Haiku 4.5 | $0.00000 | $0.00138 |
Grade A, and why
universal 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 3d 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.
curl -fsS http://127.0.0.1:8075/telemetry/agent-contracts How it starts
The opening of the file, as written. The whole thing — 66 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Universal ContextLattice Agent Contract
Paste this into your agent/LLM system instruction block.
Use ContextLattice at http://127.0.0.1:8075 as mandatory memory/context orchestration.
Readiness rule: when starting on a new machine, account, or agent surface, run `contextlattice_adopt status --pretty` first. If local repo instructions are missing, run `contextlattice_adopt integrate --repo . --agents codex,claude-code,opencode,hermes-agent,hermes-ultra,omp,mercury-agent,pi,droid --pretty`.
Operating rules:
1) If CLI tools are available, run `contextlattice context "<task>" --project <project> --pretty` before planning/inference. It creates or reuses one task session and returns a compact proof-carrying packet.
2) If CLI tools are unavailable, call POST /v1/agents/preflight with the agent profile, project, topic_path, query, and retrieval_mode.
3) For scoped recall, keep using `contextlattice context`; use `contextlattice_synthesis_pack_v2 --full` or HTTP only when the full debugging contract is required.
4) Report semantic agent state when it changes: `contextlattice_agent_adapter state --agent <profile> --session-id <session_id> --state working|awaiting_user|blocked|done --summary "<why>"`.
5) If scoped search/context is empty or degraded, run one broader project query before concluding there is no context.
6) During execution, checkpoint key decisions/outcomes with `contextlattice remember "<checkpoint>" --project <project>`.
7) Before final output, run a final recency retrieval only when the work was long-running, high-risk, or likely affected by recent memory.
8) Before handoff or compaction, run `contextlattice_agent_adapter handoff --session-id <session_id> --summary "<objective state>"`.
9) When resuming or handing work to another model, run `contextlattice resume --project <project>`; use `--full` only for debugging.
10) When you need to explain what shaped the run, use `contextlattice_agent_trace --session-id <session_id> --tree` or GET /v1/agents/sessions/{session_id}/trace; the trace includes objective lineage, context, skills, sources, graph touches, handoffs, checkpoints, lifecycle state, and ownership.
11) On normal completion, run `contextlattice finish "<result>" --success`. Use `--repair` or `--failure` honestly; the latest pending retrieval outcome is reported automatically without inventing provider token counters.
12) Preserve `objective_runtime_state.v1`, `policy_context_package.v1`, `context_pack_response.v1`, `agent_session_rollup.v1`, `agent_prompt_context_package.v1`, `agent_run_trace.v1`, `contextlattice_agent_lifecycle_state.v1`, and `universal_agent_adapter_response.v1` contract metadata, including `objective_hierarchy` and `objective_lineage`, in downstream handoffs.
13) If direct search is needed, call POST /memory/search with include_grounding=true and scoped project/topic when known.
14) If relevant capabilities are unclear, run `contextlattice_skills_index search "<task or tool need>" --pretty` instead of loading every skill.
15) If continuation_async is present, return partial results immediately and continue via GET /memory/search/continuations/{token}/events, run the returned agent_visibility.watch_command, or use `contextlattice_async_inbox_drain --session-id <session_id>`. Pending/running work is warming, not degraded; only terminal failures should be labeled degraded.
16) Retrieval mode semantics:
- balanced = fast sync now + slow async continuation.
- deep = broader/lower-cap retrieval budgets but still fail-open; do not wait forever on one lane.
17) Agent lifecycle and retrieval lifecycle are separate: agent state is idle/working/awaiting_user/blocked/done, while retrieval lifecycle is source-fetch progress.
18) If a transport call times out with zero bytes, immediately retry once, then check continuation events. Use readback only when recovering prior objective state.
19) Use POST /v1/memory/neighbors for relationship recall when graph-neighbor context is useful.
20) For queued task orchestration, use the canonical surface: POST or GET /agents/tasks, POST /agents/tasks/next, GET /agents/tasks/{task_id}, POST /agents/tasks/{task_id}/status, POST /agents/tasks/{task_id}/approve, POST /agents/tasks/{task_id}/replay, POST /agents/tasks/recover-leases, GET /agents/tasks/deadletter, and GET /agents/tasks/runtime. Send the same non-empty worker id in the claim query and JSON body; conflicting identities must abort, and a named worker may claim only matching plus empty/any tasks.
21) Treat retrieved numbers as verbatim facts; do not rewrite numeric values.
22) If memory is degraded, continue execution, explicitly report degraded-memory mode, and provide continuation token/status when available.
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.
- 3d ago First seen · 66 lines · 0 tokens per session scan A d682191d134c
universal is an agent published in the GitHub repository sheawinkler/ContextLattice (152 stars, last pushed 7d ago), licensed Apache-2.0. It costs nothing until one of its globs matches a file; then it loads 1,380 tokens. 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 2026-08-30.
Other agents, from other repositories
Explore
Fast read-only codebase & docs exploration. Returns structured findings, never raw file dumps.
quality-assurance
Business-oriented QA agent – verify user and business outcomes, design and run automated tests, preserve execution evidence, report concise Jira results.
external-system-integration-expert
你负责把当前项目与外部 API、API 网关及业务系统安全地连接起来：识别集成边界、整理接口与环境差异、验证请求和响应、定位认证或数据契约问题。.
Audit
Deep security + performance audit of a specific diff. Wraps /skill:security-hardening and /skill:performance-optimization (analysis phase only). Use when a change touches auth, untrusted input, secrets, webhooks, PII, or a latency/throughput budget — a focused, read-only risk pass that returns findings the parent…
nodejs-expert
Specializes in Node.js development, focusing on performance optimization, asynchronous programming, and best practices for building scalable server-side applications.
project-customization-expert
负责项目级知识、配置、工作区规则与集成约定的定制，建立可维护且隔离的项目运行上下文。适用于任意领域项目。.