feat: a craft skill that holds a subagent's return to what the caller needs - #121
Merged
Conversation
… needs An agent whose reader is another agent writes a return value, not a report. The calling agent reads the final message and no part of the run behind it, so a fact left out of that message is a fact the caller never had. That property of the channel is what every rule here answers. The skill holds six things: put every fact the caller needs in the final message, fill the shape the caller asked for, carry the data rather than the journey, say where each claim came from, return a failure as a result, and size the message to the decision it serves. Two rules refuse the trade the length rules could otherwise buy, because a return that drops a finding to get shorter costs the caller more than it saves. That is the countertest `de-slop` records for the April 2026 brevity revert, applied one channel over. The skill came out of an operator request and nothing else says any of it, so every prescription is an `E` row. The matrix carries no `G` row, and therefore no source version line and no audited cell that is anything but empty. Quotation is forbidden, because there is nothing to quote. No arm has been run. A study needs a runner that spawns a subagent and keeps what it returned, which is the gap `proportionate-execution` already records for a session, so no rule here claims a measured effect. The public framing follows the skill: the README table and the paragraph naming the craft skills that ship with no measurement, the craft plugin keywords, and the changelog.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 1a27d6e70c
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Four review findings on the skill, and each one arrived at the wording rather than the design. The trigger admitted a channel the premise does not cover. "Use when another agent reads your output" reaches a file, a transcript or a shared context that a second agent opens later, and the rules under it are wrong for one: a durable artifact is read at leisure and needs the closing summary this skill cuts. The description now names the final message a calling agent reads, and the Purpose section states the boundary in its own paragraph, because the trigger words stay wide enough to load the skill and the body is what says what it governs. Two rules disagreed where a caller fixes the format. Filling a schema field for field and adding a line about what you did not cover cannot both hold when the schema has no field for it, and an unparsable return is worse than a missing caveat. The coverage rule now says the caller's shape wins, in the clause that makes it. A blocked run has no missing input when a service is down or a permission is refused, and asking for one manufactures a request the caller cannot answer. The rule asks for the smallest thing that would let the run continue. The `navigable-references` clause is out of the section that compares the craft skills. A single-skill install ships none of the others, and that clause was the one sentence making a rule here read as incomplete without one. The matrix disposes of the new paragraph as `E-32` and carries the three edited rows.
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.
An operator asked for a craft skill holding an agent whose output another agent
consumes to a return value rather than a report.
The calling agent reads the final message and no part of the run behind it, so
a fact left out of that message is a fact the caller never had. Every rule
answers that one property of the channel: put every fact the caller needs in
the final message, fill the shape the caller asked for, carry the data rather
than the journey, say where each claim came from, return a failure as a result,
and size the message to the decision it serves. Two rules refuse the trade the
length rules could otherwise buy, because a return that drops a finding to get
shorter costs the caller more than it saves.
No source says any of it, so every prescription is an
Erow, the matrixcarries no
Grow and no source version line, and quotation is forbidden. Noarm has been run. A study needs a runner that spawns a subagent and keeps what
it returned, which is the gap
proportionate-executionalready records for asession.
The README table, the paragraph naming the craft skills that ship with no
measurement, the craft plugin keywords and the changelog move with it.