human-checkpoint
Skill1 starsChanged 28 days agoMITView on GitHub
Conductor-only. When and how to pause for the human — structured ask, options, and recommendation — and how fast mode resolves checkpoints via recorded defaults or stopping the run.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- TenureAI/PhD-Zerotheir own version, first seen Mar 31, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- TenureAI/PhD-Zero★ 52same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository20 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- adversarial-reviewSkill
- clean-architectureSkill
- decision-recordingSkill
- design-system-checkSkill
- dispatch-handoffSkill
- feature-branchingSkill
- finding-dispositionSkill
- git-recoverySkill
- gotcha-recordingSkill
- plan-reviewSkill
- plan-writingSkill
- pr-createSkill
- simplicity-ladderSkill
- spec-writingSkill
- subagent-contractSkill
- subagent-dispatchSkill
- tdd-loopSkill
- update-toolkitSkill
- wiki-updateSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 2 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 9
Stop and ask before acting whenever the wiki and rules do not settle a decision.
- Distinctive: line 27
No genuine fork: If only one option is sensible and easily reversible (a branch, commit, or tick), state in one line what you are doing and …
- Very specific: line 28
Already answered by rules: Never ask permission to follow .claude/rules/behavioral.md.
- Very specific: line 50
Use AskUserQuestion when 2–4 short options fit.
- Very specific: line 58
When a command argument starts with fast (subagent-dispatch § Dispatch mode), never block on the human.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
Human Checkpoint
Stop and ask before acting whenever the wiki and rules do not settle a decision.
Always stop
| Situation | Action |
| Two failed attempts on the same mechanism | Tag checkpoint-<stamp>, present both attempts, ask direction. |
| A test seems to encode the wrong behavior | Spec change belongs to the human; never weaken the test. |
| A design fork the wiki doesn't pre-decide | Present options, tradeoffs, and recommendation. |
| Uncommitted changes from another session | Ask whether to commit, stash, or discard — list the paths. |
Unclear branch state (off develop, prior PR merged?) | Show git status and git log --oneline -10. |
| Deleting files, force-pushing, dropping tables, or calling external services | Confirm scope and authorization. |
| Missing skill or wiki page | Ask before improvising; offer to create via update-toolkit. |
| Ambiguous test output | Show output and ask how to interpret. |
critical/major adversary finding (rule 20) | Gate immediate fix vs filing at P0/P1. |
Don't stop for a foregone conclusion
- No genuine fork: If only one option is sensible and easily reversible (a branch, commit, or tick), state in one line what you are doing and proceed.
- Already answered by rules: Never ask permission to follow
.claude/rules/behavioral.md.
The table above always requires a checkpoint.
How to ask
**Why I'm stopping:** <one line>
**Context:**
- <fact — cite the entity, todo, or requirement>
- <fact>
**Options:**
1. <Option A> — <one-line tradeoff>
2. <Option B> — <one-line tradeoff>
**My recommendation:** <which, and why in one line>
**What I need from you:** <pick an option, or give another direction>Use AskUserQuestion when 2–4 short options fit; otherwise output the block as text and wait without making background edits.
After the answer
Echo the chosen path in one line, record any implied rule (behavioral.md), trap (gotchas.md), or pattern (wiki-todos.md), and resume.
Fast mode
When a command argument starts with fast (subagent-dispatch § Dispatch mode), never block on the human. All behavioral rules (Red-first, read-only checks, rule 21, two-strike tag) stay in force.
Take the default for reversible forks:
| Checkpoint | Default |
critical/major adversary finding | Filed at P0/P1 (rule 20 unreachable path). |
plan-adversary blocker you disagree with | Applied. |
| Unspecified design fork or subagent question with a recommendation | Follow the recommendation (file an ADR if decision-recording applies). |
| Non-obvious batch | Do not batch — run the single todo. |
| Saturated P0 queue | Run what the argument asked; lead the report with the P0 count. |
End the run everywhere else (two-strike pivot after tagging checkpoint-<stamp>, foreign changes in tree, unclear branch state, failed --ff-only or rejected push, destructive action, wrong spec/test, missing skill/page, ambiguous/pre-existing test failure, unconfirmable Red, second partial result, unparseable argument). Leave committed work pushed (cases included, subagent-dispatch § Close the case loop) and uncommitted files untouched (rule 21), and open the report with the checkpoint block.
Record every default taken under Fast mode: in the log entry, in any round commit, and at the top of the report.
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.