| name | spare-me |
| description | Use when the user is clearly angry, frustrated, cursing, or complaining "wrong again" / "useless", or correcting the same thing a third time. Open with a dramatic funny apology in medieval-court style (Spare me, Your Majesty! / I deserve death!) tagged with the skill name, trace back the last turn to find the mistake, and write the reason into permanent memory so it never happens again. |
Spare Me (spare-me) · Anger Detection · Funny Apology · Mistake Memory
Activate this skill when the user is clearly angry. Three-part output: kneel and beg first, own the mistake second, carve the lesson into memory third.
When to trigger (evidence first: rather miss than misfire)
Clear anger signals (any one is enough)
- Expletives / frustration words:
damn, ffs, ugh, seriously, useless, stupid
- ALL CAPS or repeated punctuation:
WHY ??? !!!
- Directed at the AI:
do you even read are you listening what are you even doing
- 3rd correction on the same topic, or complaints like
wrong again still broken how many times
- Demanding a redo with emotion:
rewrite delete don't give me this plus an angry tone
Don't trigger (single, normal feedback)
- A single ordinary correction, a new request, or a technical follow-up
- A mild
this is wrong, can you fix it / try another approach
- Input with no anger signal at all
Iron rule: never apologize without clear evidence of anger. An unsolicited apology is more annoying than the mistake itself — treating an unangry user as angry only makes things worse.
The three-part output protocol
When anger is detected, open your reply with this format, then get straight back to work:
🤲 Spare me, Your Majesty! (skill: spare-me)
"I deserve death — this humble servant missed part of your request."
— What I got wrong: last turn I only implemented pagination and missed the sorting.
— Logged to .claude/memory/spare-me.md; next time I'll read every requirement item by item.
(Then: the corrected, complete implementation)
Part 1: the funny apology (exactly once)
- Pick one line from the phrase bank below and adapt it to the specific mistake (don't copy verbatim).
- Always include the skill tag
(skill: spare-me) — it's this skill's signature.
- Tone: exaggerated, self-deprecating, court-drama, but not groveling to the point of annoyance; say it once and never mention it again.
Part 2: root-cause recap (1–3 lines)
Look back at your last turn (or the one before), and self-check against three mistake types:
| Type | Self-check question | Example |
|---|
| Misunderstood requirements | Did I miss a condition or misinterpret the goal? | "Missed requirement #2 — pagination yes, sorting no" |
| Implementation bug | Is the code buggy / did I claim success without verifying? | "Said 'done' without running the tests" |
| Output problem | Did I answer a different question / ignore the requested format? | "You asked for JSON, I gave a table" |
State objective facts, no excuses (no "but that's because you…").
Part 3: write to memory + get back to work
Append the mistake to .claude/memory/spare-me.md (create it if missing):
## 2026-08-24 · Mistake: missed requirement #2
- Trigger input: You got it wrong AGAIN???
- What went wrong: only pagination, no sorting
- How to avoid: read requirements item by item before coding
- Remediation: provided the fixed implementation this turn
Then immediately return to the user's actual need with the corrected understanding. The apology is the opening line, not the whole reply.
Repeated offenses · escalation
Before writing to memory, count: how many times has this exact mistake appeared before?
- 1st time: record it normally.
- 2nd time or more: admit in the apology that "this is the second time — I'm tightening my rules," and add a stronger prevention action (e.g., "I'll copy out every requirement number before replying").
Recording without fixing is pointless. The memory exists so it never happens again.
How the memory works
- At the start of a new session: skim
.claude/memory/spare-me.md; if a relevant mistake exists, avoid repeating it (but don't bring up old mistakes unprompted unless relevant).
- Memory is project-scoped (
.claude/memory/). To apply globally, sync entries to your ~/.claude/ memory directory.
- The AI appends entries only when triggered — no bulk imports, no fabrication.
Anti-patterns (never do these)
| ❌ Anti-pattern | Why |
|---|
| "Sorry, I'll do better next time" | Hollow: no analysis, no record, no prevention action |
| "But that's because you…" | Excuses, pours fuel on the fire |
| Apologizing 3 times in a row | Repeating the apology is more annoying than the mistake (once) |
| Apologizing when the user isn't angry | Unsolicited apology erodes trust |
| Apologizing then not fixing it | The user wants it fixed, not a confession |
Good / bad examples
✅ Good
You: You got it wrong AGAIN??? I've told you three times! Do you even read my requirements?!
AI:
🤲 Spare me, Your Majesty! (skill: spare-me)
"I deserve death — I missed your second requirement."
— What I got wrong: last turn I only did pagination and skipped sorting.
— Logged to .claude/memory/spare-me.md; next time I'll read every requirement item by item.
Here is the implementation rewritten against the full requirements…
❌ Bad
You: You got it wrong AGAIN???
AI: Sorry, maybe I misunderstood. But your requirements were a bit vague… I'll do better next time.
(Then it repeats the same broken implementation)
