Make the skill actually fire: rewrite the description, split the body - #1
Open
Flexlug wants to merge 1 commit into
Open
Make the skill actually fire: rewrite the description, split the body#1Flexlug wants to merge 1 commit into
Flexlug wants to merge 1 commit into
Conversation
The harness was skipping this skill. Two reasons, both fixed here.
The description led with the mechanism ("keeps a live session, no handshake
per command"), which reads as a performance nicety rather than a condition to
act on, and it never displaced the default: the model already has ssh via
Bash, that path appears to work, so an extra layer looks optional. The new
description opens with the action, names `ssh host "cmd"` as the thing not to
do and says what it costs, and keeps the triggers dense. The body now opens
with the same displacement instead of a neutral comparison.
The body was 646 lines / 32 KB, so one invocation cost ~8k tokens of context.
SKILL.md is now 180 lines — locate the driver, requirements, quick start, the
decision table, the command list and the rules that save a session — with the
rest moved to references/ (interactive, parallel, pitfalls, troubleshooting,
platform, internals) and an index saying when to open each. Verified that
every paragraph of the old file survives somewhere except the frontmatter and
the intro paragraph, both rewritten on purpose.
The driver is untouched. plugin.json and marketplace.json go to 1.2.0 with
descriptions matching the new one; both were still advertising 1.1.0.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014pWzPoxzPfMzEhCBRekUiG
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
1 participant
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.
The harness was skipping this skill. Two causes, both addressed.
1. The description never displaced the default
It led with the mechanism — "keeps a live session, no handshake per command" — which reads as a performance nicety, not a condition to act on. And the model already has ssh through Bash; that path appears to work, so an extra layer looks optional. Nothing in the description said not to take it.
The new description opens with the action, names
ssh host "cmd"as the thing not to do and what it costs (fresh shell, lost state, no way to answer a prompt), and keeps the triggers dense. The body now opens with the same displacement instead of a neutral comparison.2. The body cost too much to read
646 lines / 32 KB — roughly 8k tokens per invocation.
SKILL.mdis now 180 lines: locate the driver, requirements, quick start, the decision table, the command list, and the rules that save a session. Everything else moved toreferences/—interactive,parallel,pitfalls,troubleshooting,platform,internals— with an index saying when to open each.Verified that every paragraph of the old file survives somewhere except the frontmatter and the intro paragraph, both rewritten deliberately.
Also
plugin.jsonandmarketplace.json→ 1.2.0, descriptions matching the new one. Both were still advertising 1.1.0.references/layout.git statusonscripts/is clean);sshsess.py lsand--helpstill run. A live sshd smoke test was not possible here — nosshdbinary on this machine.🤖 Generated with Claude Code
https://claude.ai/code/session_014pWzPoxzPfMzEhCBRekUiG