feat: add agent skills discovery instruction to llms.txt#36
Merged
Conversation
Replace the single-line skills URL in llms.txt with a structured instruction block that tells AI agents to fetch the skills index, remember skill descriptions, and load relevant skills on-demand before writing code. Closes #35
Member
Author
|
@raymondk just FYI, merging this now :) |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Replaces the single-line skills URL in
llms.txtwith a structured instruction block that tells AI agents how to discover and use ICP agent skills on-demand — without requiring any pre-installed skills.Before (buried, no instruction):
After (prominent section with clear workflow):
This mirrors the lazy-loading pattern that locally installed skills already use (name + description in context, full content loaded on-demand), but works over HTTP for any agent with zero setup.
Blind subagent test
To verify the instruction works in practice, a subagent was launched with no pre-installed ICP skills. The build was run locally to generate the updated
llms.txt, then served vianpm run previewatlocalhost:4322. The subagent was given the following prompt:How the subagent behaved:
llms.txtfrom the local preview server — found the skills section immediately (it's at the top, before the 85 doc page links)index.jsonas instructed fromskills.internetcomputer.org— discovered all 16 skills with descriptionsinternet-identity,icp-cli,canister-securityhttps://skills.internetcomputer.org/.well-known/skills/{name}/SKILL.mdic-cdk = "0.19",@icp-sdk/auth,@icp-sdk/core)ii: trueconfig for local II setup inicp.yaml@icp-sdk/auth/client,@icp-sdk/core/agent/canister-env)msg_caller()before.await, never pass caller as argument,fetchRootKeyonly in local devCargo.tomlwith version pins andcrate-type = ["cdylib"]The subagent concluded: "An agent starting cold with only the llms.txt URL can discover all of this in three HTTP requests before writing a single line of code."
This addresses the core issue from #35 — the 15 errors and 8 failed deploys were caused by the agent ignoring skills that were available to it. With the updated
llms.txt, an agent with no pre-installed skills can self-bootstrap to the same quality of guidance.Closes #35