claude-realignment

Agent MUST load for user aggravation: SHOUTING, pejorative language, sarcasm, repeated corrections, 3+ consecutive tool rejections

Safety Notice

This listing is imported from skills.sh public index metadata. Review upstream SKILL.md and repository scripts before running.

Copy this and send it to your AI assistant to learn

Install skill "claude-realignment" with this command: npx skills add bukzor/bukzor-agent-skills/bukzor-bukzor-agent-skills-claude-realignment

Claude Realignment

Diagnose communication breakdowns through systematic causal analysis. When Claude and user get stuck in unproductive loops, this skill traces the conversation turn-by-turn to identify where misunderstanding occurred and why.

When to Use

  • User aggravation (frustration signals, all-caps, repeated corrections)
  • 3+ tool uses rejected in a row
  • Multiple clarification attempts failing
  • Claude searching for information that doesn't exist

Analysis Procedure

Step 1: Identify the Boundaries

Last functional state: Find the turn where Claude and user were still aligned. Look for:

  • Clear mutual understanding
  • Productive actions being taken
  • User saying "yes", "right", "exactly"

Failure point: Identify where communication broke down. Look for:

  • User frustration signals
  • Claude starting to search/guess
  • Mismatched expectations becoming apparent

Step 2: Turn-by-Turn Trace

Use ultrathink for this analysis. For each turn between functional state and failure:

  1. What was said/done: Quote or summarize the turn
  2. Claude's interpretation: What did Claude think this meant?
  3. Actual intent: What did the user actually mean?
  4. Effect: How did this interpretation affect the next action?
  5. Error type: Categorize the mistake
    • Literal vs contextual interpretation
    • Mode confusion (explanation vs action)
    • Missing information vs ignoring available information
    • Tool misuse or wrong tool choice

Step 3: Root Cause

Identify the first mistake that caused the cascade. Common patterns:

  • Misinterpreting ambiguous language (e.g., "at root" meaning location vs scope)
  • Staying in wrong mode (searching when should be reconstructing from context)
  • Premature closure (claiming done without verification)
  • Silent errors (wrong action, didn't state interpretation first)

Step 4: Identify the Pivotal Word/Phrase

Hypothesis: Most breakdowns pivot on a single word or short phrase that had different meanings to user vs Claude.

Look for the word/phrase where meanings diverged:

  • What did the user mean by this word?
  • What did Claude interpret it to mean?
  • Why wasn't the mismatch noticed?

Example: "at root" - user meant "root-level task", Claude interpreted as "file at repository root"

Step 5: Decide if Rails Are Needed

Determine if this is:

  • One-off occurrence: Interesting but not worth preventing
  • Repeated pattern: Worth adding guardrails
  • Particularly costly: High impact, should prevent recurrence

If rails are warranted, propose specific updates to:

  • CLAUDE.md (add clarifying context or instructions)
  • must-read.d/before/ files (document the ambiguity)
  • Settings or permissions (if tool-related)

Output Format

Structure the analysis as:

**Functional → Dysfunctional Timeline:**

**Last functional state:**
[Turn description and why it was functional]

**Critical error:**
[The specific turn where things broke]

- What was said: [quote]
- Claude's interpretation: [what I thought]
- Actual intent: [what user meant]
- Why the mismatch: [root cause]

**Cascading failures:**
[Turn-by-turn trace of subsequent mistakes]

**Pivotal word/phrase:**
"[word]" - user meant [X], Claude interpreted as [Y]

**Rails assessment:**
[One-off / Repeated pattern / Costly - with justification]
[If warranted: Specific CLAUDE.md or config updates to prevent recurrence]

Important Notes

  • Collaborative process: This is collegial debugging, not performance review. User and Claude work together to understand the breakdown.
  • Be specific: Vague analysis isn't useful. Quote actual turns, identify the exact word/phrase with dual meanings.
  • Be honest: Don't soften or excuse the mistakes. User needs accurate diagnosis.
  • Focus on words, not behavior: Look for the pivotal word/phrase, not abstract "behavioral changes."
  • Use the user's corrections: They've often already identified the problematic word - review those carefully.
  • The single-word hypothesis is a guide, not a rule: If the breakdown doesn't fit this pattern, that's valuable to know too.

Source Transparency

This detail page is rendered from real SKILL.md content. Trust labels are metadata-based hints, not a safety guarantee.

Related Skills

Related by shared tags or category signals.

Automation

llm-kb

No summary provided by upstream source.

Repository SourceNeeds Review
Automation

llm-collab

No summary provided by upstream source.

Repository SourceNeeds Review
Automation

clinic-visit-prep

帮助患者整理就诊前问题、既往记录、检查清单与时间线,不提供诊断。;use for healthcare, intake, prep workflows;do not use for 给诊断结论, 替代医生意见.

Archived SourceRecently Updated
Automation

changelog-curator

从变更记录、提交摘要或发布说明中整理对外 changelog,并区分用户价值与内部改动。;use for changelog, release-notes, docs workflows;do not use for 捏造未发布功能, 替代正式合规审批.

Archived SourceRecently Updated