escalate
Skill1 starsChanged 4 months agoMITView on GitHub
Judgment escalation queue. When a session hits a judgment-dense question it shouldn't wing — a one-way door with no precedent, a borderline gate, a spec-vs-code conflict, a calibration miss — it stops that thread, queues the question to .claude/escalation-queue.md with a provisional answer, and ships the rest. Use /escalate burn to batch-adjudicate the queue with a frontier model or human.
5 other projects use a version of this.
Adopted by
Most recently taken up by
- ghvinashvili/specagenttheir own version, first seen Nov 1, 2026
- DoKu67/claude-code-skilstheir own version, first seen Oct 16, 2026
- June74/agentic-workflowtheir own version, first seen Aug 15, 2026
- Kastalien-Research/thoughtboxtheir own version, first seen Jul 3, 2026
- composio-community/support-skillstheir own version, first seen Apr 10, 2026
Other versions5 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 5
- Kastalien-Research/thoughtbox★ 64same name
- composio-community/support-skills★ 18same name
- ghvinashvili/specagent★ 7same name
- DoKu67/claude-code-skils★ 4same name
- June74/agentic-workflow★ 0same 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.
The file
We split this skill into the rules it gives an agent
4 rules. 0 are common, 2 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 5 edited versions we read keep these.
2
Distinctive
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 37
STOP the judgment thread: No further analysis; you already know it's above this session's pay grade — that's the trigger firing.
- Very specific: line 57
Read ## Queue. If empty, say so and stop.
- Distinctive: line 66
If any provisional answer was wrong AND already shipped, open a follow-up task to unwind it.
- Very specific: line 75
Don't queue two-way doors.
“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
/escalate — Queue the Hard Call, Ship the Rest
The expensive failure isn't the hard question — it's a mid-strength answer to a hard question, silently baked into shipped work. This skill splits them: the work ships now, the judgment call goes to a queue where a frontier model or a human adjudicates it in batch, with full context preserved.
Invocation
/escalate should verdicts.log move into the decisions DB? # queue a question
/escalate # queue the question under discussion
/escalate burn # batch-adjudicate the queueTriggers — queue instead of deciding inline when:
- (a) One-way door, no precedent —
/doorsays one-way and neither/precedentnor the ledger has a banked contract or prior ruling to lean on. - (b) Borderline gate — a
/gate checklands within ~10% of threshold in either direction. Don't round; queue. - (c) Spec-vs-code conflict —
/driftfinds a disagreement and neither side is clearly authoritative. - (d) Systematic error —
/calibratesurfaces a pattern whose correction isn't obvious (which mechanism to change, or whether to demote a domain).
Queue procedure
- STOP the judgment thread. No further analysis; you already know it's above this session's pay grade — that's the trigger firing.
- Append an entry to
.claude/escalation-queue.md(create with## Queueand## Adjudicatedsections on first use; commit it — the queue is the point):
IDs are sequential (grep the file for the last ESC-).### ESC-NNN | 2026-07-09 | trigger: one-way-no-precedent Q: <the question, one or two sentences> Context: <files, claims, numbers — everything an adjudicator needs cold> Provisional: <this session's own answer> [MED] Shipped despite: <what work went out with this question open> - Take the provisional path that keeps the door open — prefer the option
with the cheapest escape hatch, add the seam if you can (see
/doorstep 4). - Ship the rest of the work. The queue entry is the record that the open question was seen, not skipped.
- Verdict line:
ESCALATE: ESC-NNN | <trigger> | queued
Burn procedure (/escalate burn)
- Read
## Queue. If empty, say so and stop. - For each entry, adjudicate with the strongest available judge — frontier model session or human. Present the full entry; get a ruling and a one-line rationale. The provisional answer is scored (right/wrong) — free calibration data.
- Log each ruling:
/verdict log ESCALATE: ESC-NNN | <trigger> | adjudicated — <ruling>. - Move the entry from
## Queueto## Adjudicated, appending the ruling, rationale, adjudicator, and date. Never delete — adjudicated entries are the precedent store/precedentgreps. - If any provisional answer was wrong AND already shipped, open a follow-up task to unwind it; note it in the entry.
Rules
- Queueing is not punting: an entry without a provisional answer + confidence tag is rejected. You must commit to a best guess so the burn can score it.
- One question per entry. A tangle of questions means you haven't isolated the judgment call yet — do that first.
- Don't queue two-way doors. If undoing it is cheap, just pick (see
/door). - A queue older than one burn cycle with unshipped entries is a smell: either burn it or admit the questions didn't matter and close them as such.
Composes with
/door— trigger (a); one-way doors with no precedent land here instead of being decided inline. The disagreement check in /door routes here too./gate— trigger (b); borderline results queue rather than round./drift— trigger (c); unadjudicatable spec-vs-code conflicts queue here./calibrate— trigger (d); and the burn's provisional-vs-ruling scores feed the next calibration round./verdict— every adjudication is logged there; the queue's Adjudicated section is a/precedentsource.
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.

