b0f7c8640b52cdee33c452dffc702daec53b9bc7
3
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Evan
|
102b506b0b |
fix(agents,hands): per-agent/per-hand mcp_servers / skills allowlists (#87) (#92)
All 32 agent manifests and 17 hands shipped with empty mcp_servers /
skills lists, which the kernel interprets as "no filter" — every
globally-configured MCP server's tools and every installed skill get
injected into the prompt on every LLM call. On a typical instance (9
MCP servers, ~85 MCP tools + ~82 built-in tools) that's ~50k input
tokens per turn spent on definitions the agent never uses.
Changes
-------
32 agents/*/agent.toml:
- mcp_servers: 1-4 per agent. memory wherever state persists across
turns; fetch / exa-search / brave-search only where the prompt
actually calls for web; git / github / filesystem on engineering
agents; gmail / google-calendar / linear / jira on productivity
agents whose prompts mention them.
- skills: per-role allowlist driven by what the system_prompt names
(e.g. coder → rust/python/typescript/git/shell-scripting; devops-
lead → docker/kubernetes/terraform/ansible/ci-cd/helm/prometheus/
sysadmin). Generalists (assistant) keep skills = [] (see "Open
items" below).
- skills_disabled = true on the four short-conversational agents
(hello-world, recipe-assistant, health-tracker, home-automation).
Their system prompts never instruct the LLM to consult any skill,
so loading all 60 was pure waste. They also drop the explicit
max_history_messages override and inherit the kernel default (60).
- max_history_messages tiered by workload shape:
60 short conversational (hello-world, recipe, health-tracker,
home-automation) — inherits the rising kernel default
(`DEFAULT_MAX_HISTORY_MESSAGES = 60`); no override needed.
60 single-turn task agents (writer, translator, doc-writer,
email-assistant, customer-support, sales-assistant, recruit-
er, social-media, personal-finance, tutor, travel-planner,
meeting-assistant, ops, devops-lead, planner) — explicit
override at the same value to lock the cap if the kernel
default moves again.
80 multi-step / tool-heavy (coder, debugger, architect, code-
reviewer, test-engineer, security-auditor, analyst, data-
scientist, academic-researcher, researcher, legal-assistant)
120 coordinators (assistant, orchestrator) — long multi-agent
sessions where prompt-cache continuity is critical
All values sit at or above the kernel default. Pinning lower
would thrash the prompt cache (the failure mode #91 fixed for
the creator hand by *raising* the cap, not lowering it).
17 hands/*/HAND.toml:
- hand-level mcp_servers / skills now declared on every hand, so
every [agents.*] inside inherits a sensible allowlist.
- skills_disabled = true placed on each [agents.*] inside clip and
creator (pure media pipelines that don't benefit from any skill).
HandDefinitionRaw in librefang-hands does NOT have a top-level
skills_disabled field — declaring it at the hand top level would
be silently dropped by serde, so the setting must live on the
AgentManifest of each sub-agent role.
- devteam: expand existing mcp_servers = ["github"] to include
memory / git / filesystem; populate skills with the expected
dev-team expertise (replacing the placeholder skills = []).
- wiki: replace placeholder mcp_servers = [] with [memory, fetch,
filesystem]. Hand-level skills stays [].
- lead: hand-level skills was originally [email-writer, writing-
coach, interview-prep]; interview-prep is for job-interview
preparation, not lead generation. Replaced with data-analyst
(used by the qualification-scoring step in the prompt).
schema.toml: register mcp_servers / skills / max_history_messages on
the agent field schema so machine consumers (RegistrySchema in
librefang-types) see the new top-level fields. The
max_history_messages description now points at
librefang_runtime::agent_loop::DEFAULT_MAX_HISTORY_MESSAGES (60
today) by name, so the schema doesn't go stale when the constant
moves again.
agents/README.md: example block + "Adding a New Agent" checklist
mention the allowlists; max_history_messages example is shown
commented out with a prompt-cache caveat.
Open items
----------
`assistant` (the default user-facing agent) keeps `skills = []`
deliberately. It is the generalist entry point — capping its skill
surface at a small allowlist would defeat its "delegate to any
specialist" job. The trade-off is that this single agent still pays
the full skill-definition load on every turn; operators who want a
strict allowlist for `assistant` can override it after install.
Why not adopt PR #89's approach
-------------------------------
#89 covers similar ground but with three issues this PR avoids:
1. mcp_servers = ["_none"] sentinel. #89's body explicitly notes
it's pending upstream librefang#4808 (mcp_disabled). Shipping a
magic-string today means coming back later to clean it up. This
PR uses real allowlists.
2. max_history_messages = 8 / 12 / 15 / 20. Far below today's
kernel default (60) and #91's direction for long-workflow hands
(80–120). Every turn that hits the cap invalidates the cached
prompt prefix; the cost of cache misses exceeds the saving from
shorter history. This PR uses 60–120.
3. Doubling max_llm_tokens_per_hour (coder 200k→500k, assistant
300k→500k) widens the per-agent budget — the opposite direction
from #87's "reduce per-call cost" goal. Left to the operator's
instance-specific tuning.
Refs librefang/librefang-registry#87, librefang/librefang-registry#89
|
||
|
Evan
|
d43077afa9 |
fix(providers): remove ~anthropic, skip ~ prefixes in sync script (#69)
* fix(providers): remove ~anthropic, skip ~ prefixes in sync script
OpenRouter uses ~ prefixes for internal auto-routing aliases (e.g. ~anthropic).
These are not real providers — they already route through openrouter.toml.
The generated ~anthropic.toml was confusing (looked like a stale backup)
and redundant with the existing openrouter provider.
- Delete providers/~anthropic.toml
- Skip provider IDs starting with ~ in sync-pricing.py --create-missing
* fix(providers): remove morph, aider, kwaipilot
- morph: specialized code-editing/patching tool, not a general LLM provider
- aider: CLI meta-tool wrapper (base_url empty), redundant with claude-code/codex-cli/gemini-cli/qwen-code
- kwaipilot: Kwai internal coding assistant routed via OpenRouter, niche
* fix(sync): add morph/aider/kwaipilot to SKIP_PROVIDERS to prevent re-creation
* feat(sync): merge OpenRouter-only providers into openrouter.toml
Instead of generating standalone .toml files that just wrap the OpenRouter
endpoint, merge their models directly into openrouter.toml with the
standard 'openrouter/{provider}/{model}' ID convention.
- Add _build_model_fields() and _model_lines() helpers to deduplicate
model rendering between standalone and merged paths
- Add merge_into_openrouter() that appends new models idempotently
- generate_provider_toml() now only runs for providers in PROVIDER_API
- --create-missing routes OpenRouter-only providers to merge_into_openrouter
* fix(providers): remove 14 OpenRouter-only standalone files
These providers have no direct public API and all route through
openrouter.ai/api/v1. Per the new sync-pricing.py policy, their models
will be merged into openrouter.toml on the next CI run instead of
living in separate files that just wrap the OpenRouter endpoint.
Removed: allenai, deepcogito, essentialai, inclusionai, inflection,
liquid, meituan, nex-agi, nousresearch, prime-intellect, relace,
switchpoint, tngtech, writer
* fix(providers): remove 7 niche providers with no driver support
No dedicated LLM driver code exists for these providers — they rely
purely on OpenAI-compatible passthrough with no special handling.
Removing them reduces registry noise; users can still reach them via
openrouter.toml if needed.
Removed: microsoft, ibm-granite, xiaomi, upstage, inception, aion-labs, arcee-ai
* fix(providers): remove ai21, chutes, venice
All three use ApiFormat::OpenAI with no special handling — pure passthrough.
No registry entry needed; users can reach them via openrouter.toml or by
adding a custom provider.
* docs(providers): rewrite README with full provider catalog and inclusion criteria
- List all 46 providers grouped by category with descriptions
- Document why each provider exists (direct API, unique endpoint, dedicated driver, local, CLI)
- Add inclusion criteria section explaining when to create standalone files vs merging into openrouter.toml
- Document sync script routing logic
- Update model counts: 49→46 providers, 339→232 models
* docs: add comprehensive READMEs for all registry sections + deepinfra provider
- agents/README.md: 32 agents across 7 categories with capability field reference
- channels/README.md: 44 channels across 5 categories with protocol reference table
- hands/README.md: 18 hands across 5 categories with HAND.toml format guide
- mcp/README.md: 33 MCP servers across 5 categories with transport/auth format
- plugins/README.md: 12 plugins with hook protocol documentation
- skills/README.md: 60 skills across 9 categories with SKILL.md format guide
- providers/deepinfra.toml: add DeepInfra serverless inference (5 models)
|
||
|
Evan Hu
|
206169c1d7 |
docs: add README for every content directory
Each directory (agents, hands, integrations, plugins, providers, scripts, skills) now has a README documenting its TOML format, current contents, and contribution steps. |