fft-host-ipc
Skill4 starsChanged 8 months agoMITView on GitHub
Host IPC ABI for outbound messages, scheduler control, memory/skill actions, subagents, and file delivery. Use when writing JSON into the group IPC directories or when unsure of payload shapes.
In this repository23 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- add-telegramSkill
- add-telegram-swarmSkill
- agent-browserSkill
- agent-harness-engineerSkill
- autoresearch-createSkill
- autoresearch-finalizeSkill
- fft-coder-opsSkill
- fft-dashboard-opsSkill
- fft-debugSkill
- fft-farm-bootstrapSkill
- fft-farm-onboardingSkill
- fft-farm-opsSkill
- fft-setupSkill
- fft-telegram-opsSkill
- loopsSkill
- rapid-researchSkill
- skill-mechanicSkill
- web-searchSkill
- customizeSkill
- debugSkill
- setupSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 4 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 36
Run progress (no separate chat bubble):
- Very specific: line 44
Read current_tasks.json when needed.
- Distinctive: line 72
Use skill_list/skill_view before reinventing a workflow.
- Distinctive: line 131
Never run destructive git commands unless the operator explicitly requests them.
- Very specific: line 134
Do not invent new top-level IPC type strings.
- Distinctive: line 135
Never claim delivery/mutation success without a success result file.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
FFT Host IPC
The agent talks to the host only through filesystem IPC under the run's IPC
directory (see system prompt Workspace section for the path). Write atomically
(temp file then rename). Wait for action_results/<requestId>.json with
status=success before reporting completion for request/response actions.
When to use this skill
- Sending proactive chat messages or run_progress updates
- Pausing/resuming/cancelling scheduled tasks or registering groups
- memory_search / memory_get / memory_write
- skill_list / skill_view / skill_create / skill_patch / skill_rollback
- Spawning a subagent
- Delivering files/photos back to Telegram
When not to use this skill
- Ordinary file edits in the workspace (use tools directly)
- Reading SOUL/NANO/TODOS/MEMORY (already injected or on disk)
- Coding work that belongs in fft-coder-ops
Messaging (messages/*.json)
Proactive message:
{"type":"message","chatJid":"<jid>","text":"<text>"}Run progress (no separate chat bubble):
{"type":"run_progress","chatJid":"<jid>","requestId":"<current request_id>","text":"Run status: ...","phase":"thinking|tool_running|stale","detail":"..."}Scheduler (current_tasks.json + task JSON)
Read current_tasks.json when needed. Control tasks with:
{"type":"pause_task","taskId":"..."}{"type":"resume_task","taskId":"..."}{"type":"cancel_task","taskId":"..."}- Main-only:
{"type":"refresh_groups"} - Main-only:
{"type":"register_group","jid":"...","name":"...","folder":"...","trigger":"@AssistantName"}
Task management scheduling is host-owned; prefer host/task surfaces over inventing new schedule formats.
Memory actions (actions/*.json → action_results/<requestId>.json)
{"type":"memory_action","action":"memory_search","requestId":"<id>","params":{"query":"...","topK":8,"sources":"all"}}{"type":"memory_action","action":"memory_get","requestId":"<id>","params":{"path":"MEMORY.md"}}{"type":"memory_action","action":"memory_write","requestId":"<id>","params":{"intent":"todo_upsert_task","payload":{"entryId":"T1","text":"...","status":"PENDING"}}}For writes, wait for status=success before telling the user it stuck.
Skill actions (actions/*.json → action_results/<requestId>.json)
Use skill_list/skill_view before reinventing a workflow. Mutations are host-gated to agent-created runtime skills; repo and personal source skills are read-only.
{"type":"skill_action","action":"skill_list","requestId":"<id>","params":{"includeArchived":false}}{"type":"skill_action","action":"skill_view","requestId":"<id>","params":{"name":"skill-name"}}{"type":"skill_action","action":"skill_create","requestId":"<id>","params":{"name":"short-skill-name","description":"When to use...","content":"---\nname: short-skill-name\ndescription: ...\n---\n\n# ..."}}{"type":"skill_action","action":"skill_patch","requestId":"<id>","params":{"name":"skill-name","content":"complete replacement SKILL.md"}}{"type":"skill_action","action":"skill_write_file","requestId":"<id>","params":{"name":"skill-name","filePath":"references/example.md","fileContent":"..."}}{"type":"skill_action","action":"skill_rollback","requestId":"<id>","params":{"name":"skill-name"}}Subagent (actions/*.json → action_results/<requestId>.json)
{"type":"subagent_action","action":"spawn_subagent","requestId":"<id>","params":{"task":"clear, complete instructions for the subagent","mode":"execute"}}mode:"execute"(default, full tools) or"plan"(read-only)- Poll
action_results/<requestId>.jsonuntilstatus=success - Subagents cannot spawn subagents
File delivery (deliver_files/*.json)
{
"type": "farm_action",
"action": "deliver_file",
"requestId": "<unique-id>",
"params": {
"filePath": "path/to/file.jpg",
"caption": "Optional caption text",
"kind": "photo"
}
}filePath: absolute or relative to group workspacekind:photo|document|video|audio(auto-detected if omitted)chatJid: optional; defaults to the group's registered chat- Confirm
action_results/<requestId>.jsonhasstatus=successbefore reporting delivered
Guardrails
- Never run destructive git commands unless the operator explicitly requests them.
- Preserve unrelated worktree changes.
- Atomic write only (temp + rename).
- Do not invent new top-level IPC
typestrings; host kernel freezes the ABI. - Never claim delivery/mutation success without a success result file.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

