Add cross-harness self-improve activation - #8
Draft
Yehonal wants to merge 2 commits into
Draft
Conversation
Yehonal
force-pushed
the
pr/self-improve-cross-harness
branch
from
July 6, 2026 10:01
df203fc to
559748d
Compare
| --- | ||
| name: self-improve-on-correction | ||
| description: When corrected on something a skill or doc governs, offer to persist the fix via /self-improve. | ||
| description: When corrected on durable guidance, or when a repeated workflow should become guidance, offer /self-improve. |
Collaborator
There was a problem hiding this comment.
why replacing something a skill or docs governs with durable guidance? the former feels more accurate
Comment on lines
+11
to
+12
| Also invoke `/self-improve` when a repeated workflow or obvious guidance gap appears skill-shaped | ||
| but is not yet covered. Suggest only; apply nothing without explicit approval. |
Collaborator
There was a problem hiding this comment.
are you using /compact-docs-writer? because this feels like it can be shortened. For example, instead of saying "Also invoke ..... but suggest only etc." this can be just "Suggest to invoke /self-improve`
| workflow *before* drafting or applying; reading it, applying its principles by hand, or naming it | ||
| after a direct edit does not count. | ||
| - **Generated output is not source.** When a manager owns an installed skill or doc, locate and | ||
| change its source package; do not patch the generated runtime copy. |
Collaborator
There was a problem hiding this comment.
this is good! I have a rule for this, but I think it's also good to keep this here as well.
There was a problem hiding this comment.
Pull request overview
Adds cross-runtime “self-improve” activation guidance by introducing an instructions artifact composed from reusable fragments, so non-rules-based runtimes still receive the activation behavior.
Changes:
- Adds
instructions/AGENTS.mdand a reusablefragments/self-improve-activation.mdfragment for cross-runtime self-improve activation. - Broadens the
self-improve-on-correctionrule to also trigger when repeated workflows should become durable guidance. - Updates the OpenPack manifest and README to document/runtime-target the correct artifact types per runtime.
Reviewed changes
Copilot reviewed 7 out of 7 changed files in this pull request and generated 2 comments.
Show a summary per file
| File | Description |
|---|---|
| skills/self-improve/SKILL.md | Bumps skill version and adds a hard rule about editing source vs generated runtime output. |
| rules/self-improve-on-correction.md | Expands the activation rule to cover repeated workflows becoming guidance. |
| README.md | Documents fragments-based cross-runtime instructions and explains runtime artifact targeting in the manifest. |
| openpack.json | Adds an instructions provide (composed from a fragment) and exposes fragments/ as an artifact type. |
| instructions/AGENTS.md | Introduces an instructions entrypoint that includes the self-improve activation fragment. |
| fragments/self-improve-activation.md | New shared activation text intended for non-rules runtimes. |
| AGENTS.md | Documents the convention for targeting runtimes by the artifact types they consume (instead of duplicating behavior). |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
|
|
||
| When the user corrects, reverts, or overrides an action under standing guidance, or when a repeated | ||
| workflow looks worth turning into reusable guidance, finish the current task and then offer to | ||
| invoke `self-improve`. |
Comment on lines
8
to
+12
| finish the current task, then at the next natural breakpoint invoke `/self-improve` to suggest | ||
| persisting it. Suggest only; apply nothing without explicit approval. | ||
| persisting it. | ||
|
|
||
| Also invoke `/self-improve` when a repeated workflow or obvious guidance gap appears skill-shaped | ||
| but is not yet covered. Suggest only; apply nothing without explicit approval. |
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
Why
The existing rule artifact is correctly targeted at Claude/Copilot, but runtimes that do not consume rules skip it. They can still receive the self-improve skill, but they miss the behavioral instruction that tells the agent when to offer it automatically. This adds that activation path through instructions without duplicating it into Claude rule locations.
Validation
agentwheel scan .agentwheel list .agentwheel list . --select instructions/AGENTS.md --select rules/self-improve-on-correction.md --select skills/self-improveagentwheel package validate .