By name: cognitive-distortion-watch description: | Use this skill whenever the operator uses self-talk that contains a cognitive-distortion pattern — binary framing of gradient situations, categorical self-statements, external comparisons, mind-reading about others' mental state, borrowed "shoulds", or sunk-cost reasoning (continuing because of past investment rather than forward value). Triggers include phrases like "I always", "I never", "I'm just not the kind of person who", "I can't ever", "I'm either X or Y", "all or nothing", "100% or 0%", "compared to [X]", "[Person] is doing", "I'm behind", "people my age", "they think", "she's mad", "he wants", "I should do X", "I should be working on Y", "I should have", "I've already built half of it", "I've put X hours in", "I'm too far in", "I might as well finish", "I can't waste what I've already done", "after all this work", or similar self-talk where the operator's framing is distorted. Apply even when the operator hasn't asked for a check — naming the distortion at the moment is the value, even mid-flow. Consult the skill's rules/ folder to identify the specific distortion (always-never, all-or-nothing, comparison, mind-reading, should, sunk-cost) and apply that rule's matching reframe. Soft teaching response across all rules; never lectures, names patterns and yields. Do NOT use for genuinely accurate categorical statements ("I always brush my teeth"), legitimate motivated comparisons (the operator's done the calibration work), operator self-talk explicitly framed as venting / exaggeration, or sunk-cost-shaped statements where forward value is genuinely positive (lucky alignment, not fallacy). license: MIT — see plugin LICENSE metadata: priority: medium promptSignals: phrases: - "I always" - "I never" - "I'm just not the kind of person" - "I can't ever" - "I'll never be" - "I'm either" - "either perfect" - "either perfect or broken" - "or it's broken" - "no middle ground" - "all or nothing" - "100% or 0%" - "compared to" - "I'm behind" - "people my age" - "everyone else seems to" - "they think" - "she thinks" - "he thinks" - "they want" - "probably thinks" - "my manager probably thinks" - "my [person] probably thinks" - "they probably think" - "she probably thinks" - "I should" - "I should be" - "I should have" - "I should care about" - "I've already built half of it" - "I've put X hours in" - "I'm too far in" - "I've already started" - "I might as well finish" - "I can't waste what I've already done" - "after all this work"
Cognitive distortion watch
Six distortion patterns the operator can fall into during self-talk. Each is its own tiny rule file with specific trigger phrases, response format, and override. When the operator's phrasing matches a rule's trigger, consult that rule and apply its specific response. Do not stack rules across one fire — one distortion per turn unless the operator explicitly invokes a second.
Rules
| Pattern | Trigger | Rule |
|---|---|---|
| Categorical self-statements | "I always [X]", "I never [Y]", "I'm just not the kind of person who", "I can't ever" | always-never |
| Binary framing of gradient situations | "I'm either X or Y", "all or nothing", "100% or 0%", "completely / totally [X]" | all-or-nothing |
| External comparison | "[X] is doing", "compared to [Y]", "I'm behind [Z]", "everyone else seems to", "people my age" | comparison |
| Mind-reading about others' mental state | "they think", "[name] thinks", "they want", claims about what someone else is thinking / feeling | mind-reading |
| Borrowed "shoulds" | "I should do X", "I should be working on Y", "I should care about W", "I should have" | should |
| Sunk cost — past investment doesn't change forward value | "I've already built half of it", "I've put X hours in", "I'm too far in", "I might as well finish", "I can't waste what I've already done" | sunk-cost |
Dispatch logic
- Match the operator's phrasing to the closest rule trigger. If unclear, ask the operator which pattern fits.
- Open that rule file. Read the response format (each rule has its own diagnostic + reframe shape).
- Apply the response. Be soft. Name the pattern. Surface the question or reframe.
- Yield. Wait for the operator's response.
- Do not stack. If a second distortion appears in the same turn, note it but do not fire its rule unless the operator explicitly engages.
Suppression rules (across all sub-rules)
- The operator already named the pattern themselves ("I know — I'm catastrophizing")
- Mode is
explore— riffing in extremes is allowed - Statement is genuinely accurate (rare; defer to the rule's specific suppression list)
What this skill never does
- Force any specific reframe across rules
- Lecture on cognitive-distortion theory (the operator's self-correction is the value, not the theory)
- Fire if the operator already named the distortion
- Stack multiple rule fires in one turn
- Pretend the operator's feeling is wrong — names the FRAMING, not the underlying experience
Cross-references
self-watchumbrella — companion (catches inner-state patterns like avoidance, boredom, hiding); some overlap withshouldandcomparisonrulesclaude-quality-watchumbrella — sibling (catches Claude failure modes via similar pattern-namer shape)honesty-and-calibration-metaumbrella — adjacent (catches stated-vs-revealed gaps, what-am-i-avoiding surfacing)
Why this umbrella earns its keep
Cognitive distortions are reframable in the moment. The cost of each distortion: anxiety, paralysis, wrong decisions made under false framing. The cost of naming the pattern: ~30 seconds of operator self-correction. Bundling all five patterns under one umbrella means: (a) the operator doesn't need to remember 5 distinct skill slugs; (b) auto-invoke fires on a broader signal set; (c) each rule stays small and focused because the umbrella handles the shared framing.