claude-full-send-guard
Skill0 starsChanged 57 days agoMITView on GitHub
Claude Code hook that stops a full-send program from ending the turn while units are unfinished and nothing is running.
In this repository56 other files it gives its agents.
- AGENTS.mdAGENTS.md
- instructions/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- instructions/CLAUDE.mdCLAUDE.md
- author-skillSkill
- architectSkill
- arenaSkill
- astack-helpSkill
- astack-updateSkill
- brag-slimSkill
- bug-fixSkill
- claude-background-jobsSkill
- claude-merge-gateSkill
- claude-restore-skillsSkill
- claude-routerSkill
- codex-background-jobsSkill
- correctSkill
- councilSkill
- create-verification-skillSkill
- diagnosing-bugsSkill
- domain-modelingSkill
- full-sendSkill
- grillingSkill
- handoffSkill
- harden-prSkill
- hillclimb-councilSkill
- hillclimbSkill
- howSkill
- implementSkill
- improve-codebase-architectureSkill
- investigationSkill
- multi-phase-planSkill
- no-commentsSkill
- open-prSkill
- orchestrateSkill
- perf-issueSkill
- principlesSkill
- prototypeSkill
- refactoringSkill
- reflectSkill
- review-prSkill
- run-jobSkill
- ship-prSkill
- show-me-your-workSkill
- t3-previewSkill
- tddSkill
- teachSkill
- technical-writingSkill
- to-specSkill
- typescript-best-practicesSkill
- unslopSkill
- wayfinderSkill
- whySkill
- wizardSkill
- worktree-cleanupSkill
- writing-for-agentsSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 20
No unit in units.tsv is running.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Keep a full-send program moving
scripts/hook.py is a Claude Code Stop hook. It acts when the session's history
holds the user's /full-send command, including history that a broken resume
dropped from context. After that command, full-send posts
the absolute path of the store's assumed-decisions.md. The hook reads the
program's store from the first such path in the coordinator's own text, and
ignores paths in tool output. It does nothing for a store without a units.tsv,
such as the store of a task on full-send's implement route, including one that
grows a second PR.
The hook blocks the end of the turn when all of these hold:
- No unit in
units.tsvisrunning. - Some unit is unfinished. A unit is finished when it is
done,abandoned, orzombie-reconciled. While a gate is open, ablockedunit also counts as finished. - The latest dispatch order in
preferences.mdis notDispatch paused. - Claude Code reports no background task or scheduled wakeup in the session.
The block names the store, the unfinished units, and the open gates. The model
either starts ready work or records in the store why the program waits. Then it
ends the turn. Claude Code marks that second stop with stop_hook_active, and
the hook lets it end. A later turn that ends idle is blocked again.
Unit dependencies live in the plan, so the hook cannot tell a ready unit from a
unit that waits on another unit or a gate. The model judges readiness, and it
judges whether "no merge" finishes a merge-ready unit. Each of these cases
costs one blocked stop per turn end:
- A gate holds a
needs-verifyormerge-readyunit, which cannot be setblocked. - A
blockedunit waits on something without an open gate. - The session waits on a PR watch, a T3 Code delegate, or a promised message
from another thread. Claude Code does not report work outside the session.
The coordinator avoids the delegate case by setting the unit
running. - A program under "no merge" has finished.
Gates do not name their units. So while any gate is open, every blocked unit
counts as finished, including one whose own gate was answered. A running unit
whose worker has stopped also keeps the hook from blocking, and so does a
background task that never exits, such as a dev server. When the hook cannot read
the transcript or the store, it lets the turn end.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

