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/marlburrow/hivekeep/system-promptsgit clone --depth 1 https://github.com/MarlBurroW/hivekeepWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/marlburrow/hivekeep/system-prompts)<a href="https://agentmods.dev/agents/marlburrow/hivekeep/system-prompts"><img src="https://agentmods.dev/badge/agents/marlburrow/hivekeep/system-prompts.svg" alt="Measured on agentmods" height="20"></a>What 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 | $0.00014 | $0.01146 |
| Opus 5 | $0.00007 | $0.00573 |
| Sonnet 5 | $0.00003 | $0.00229 |
| Haiku 4.5 | $0.00001 | $0.00115 |
Grade A, and why
system-prompts scanned grade A with 0 findings 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 4d 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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 78 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Every Agent's behavior is shaped by its system prompt, which Hivekeep assembles automatically from several sources. Understanding this helps you write better Agent configurations.
Prompt architecture
Hivekeep builds the system prompt from these blocks (in order):
- Platform context: explains that the Agent lives on Hivekeep, has a continuous session, and sees multiple users
- Identity: name, slug, and role
- Core principles: universal baseline behaviors (genuine helpfulness, resourcefulness, privacy, calibrated responses, tool-call discipline). Includes instructions to never narrate or predict tool results before execution, and to never batch dependent tool calls, each of which must be called one at a time across separate steps. Injected for all main Agents, not sub-Agents or quick sessions
- Personality: the
characterfield you define - Expertise: the
expertisefield you define - Platform directives: optional global prompt that applies to all Agents (set in Settings)
- Contacts directory: shared contacts across the platform
- Agent directory: other Agents available for collaboration, with delegation instructions (Hub Agents get an enriched view with expertise summaries)
- Your memory: the Agent's curated memory profile, always present. The episodic archive is not injected; the Agent searches it with
recallwhen it needs it - Internal instructions: tool usage guidelines, memory management, contact resolution, secrets handling, response calibration, mini-app creation. Includes a file & code tool selection table that steers Agents toward structured tools (
grep,multi_edit,edit_file) overrun_shellfor file operations. Mini-app instructions direct Agents to callget_mini_app_docsfor the full SDK reference rather than embedding it inline. MCP tool sections show server-level summaries only (individual tool descriptions are provided via the LLM'stoolsparameter). Channel instructions include guidance on usingattach_file()to send files back to external platforms - Workspace: when an Agent has a workspace directory configured, shows the absolute path and a file tree of its contents. Instructs the Agent to use the workspace for all file operations (repos, scripts, data) and avoid writing to the home folder or other system paths. Empty workspaces get a hint to start organizing
- Current speaker profile: name, role, and contact notes (both global/shared and per-Agent private notes) for the user who sent the current message. If the user has a linked contact but no notes yet, includes a gentle nudge to discover them naturally. Also resolves channel senders (Telegram, Discord, WhatsApp) to their contact records via platform ID
- Channel origin context: when the current turn is part of a causal chain originating from an external channel (e.g. inter-Agent reply or task result), informs the Agent that delivery is automatic and advises adapting formatting for the target platform
- Language: response language from the user's Agent language setting in account settings (falls back to the interface language when unset). Agents can speak nearly any language, independent of the UI translation.
- Date and context: current timestamp
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.
- 4d ago First seen · 78 lines · 14 tokens per session scan A f996d33702fc
system-prompts is an agent published in the GitHub repository MarlBurroW/hivekeep (51 stars, last pushed today), licensed MIT. It adds 14 tokens to every session and 1,146 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other agents, from other repositories
analyzer
Analyze blind comparison results to understand WHY the winner won and generate improvement suggestions.
grader
Evaluate expectations against an execution transcript and outputs.
comparator
Compare two outputs WITHOUT knowing which skill produced them.
AGENTS
Agents extend SidebarAgent (in base.py). The base class owns the LLM-in-a-loop mechanics: LLM init, tool schema assembly (always includes sidebartool), input sanitization (when sanitizeuntrustedinput=True), message loop with heartbeat, tool execution, completetask detection, trace capture, and completion publishing…
analyzer
Analyze blind comparison results to understand WHY the winner won and generate improvement suggestions.
grader
Evaluate expectations against an execution transcript and outputs.