agentleFS
ronniepinnellronniepinnell/casper/skills/escalate/SKILL.md

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

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
  1. Distinctive:

    STOP the judgment thread: No further analysis; you already know it's above this session's pay grade — that's the trigger firing.

    line 37
  2. Very specific:

    Read ## Queue. If empty, say so and stop.

    line 57
  3. Distinctive:

    If any provisional answer was wrong AND already shipped, open a follow-up task to unwind it.

    line 66
  4. Very specific:

    Don't queue two-way doors.

    line 75

“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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/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 queue

Triggers — queue instead of deciding inline when:

  • (a) One-way door, no precedent — /door says one-way and neither /precedent nor the ledger has a banked contract or prior ruling to lean on.
  • (b) Borderline gate — a /gate check lands within ~10% of threshold in either direction. Don't round; queue.
  • (c) Spec-vs-code conflict — /drift finds a disagreement and neither side is clearly authoritative.
  • (d) Systematic error — /calibrate surfaces a pattern whose correction isn't obvious (which mechanism to change, or whether to demote a domain).

Queue procedure

  1. STOP the judgment thread. No further analysis; you already know it's above this session's pay grade — that's the trigger firing.
  2. Append an entry to .claude/escalation-queue.md (create with ## Queue and ## Adjudicated sections on first use; commit it — the queue is the point):
    ### 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>
    IDs are sequential (grep the file for the last ESC-).
  3. Take the provisional path that keeps the door open — prefer the option with the cheapest escape hatch, add the seam if you can (see /door step 4).
  4. Ship the rest of the work. The queue entry is the record that the open question was seen, not skipped.
  5. Verdict line: ESCALATE: ESC-NNN | <trigger> | queued

Burn procedure (/escalate burn)

  1. Read ## Queue. If empty, say so and stop.
  2. 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.
  3. Log each ruling: /verdict log ESCALATE: ESC-NNN | <trigger> | adjudicated — <ruling>.
  4. Move the entry from ## Queue to ## Adjudicated, appending the ruling, rationale, adjudicator, and date. Never delete — adjudicated entries are the precedent store /precedent greps.
  5. 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 /precedent source.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.