rejection-sensitivity-check

Calibrate a message that stung — separate evidence from interpretation, score how much sting is warranted. Use when the user says "did they mean it that way", "is this criticism", "am I reading this w

SKILL.md

GitHub

Rejection Sensitivity Check

A message landed harder than expected. The user wants to know: is this actually critical, or am I reading hostility into something neutral? This skill separates the evidence in the text from the interpretation the brain layered on top.

Sibling to break-it-down. Break-it-down is general decoding; this one is specifically for messages that triggered a rejection-sensitive reaction.

Lens: psychologist + devils-advocate

This skill always invokes clarity-council via Skill in multi mode with personas [psychologist, devils-advocate]. The two-persona setup is the whole point — psychologist reads the text neutrally, devils-advocate stress-tests both the worst-case and the best-case reading.

  • user_problem: "The user received this message and it stung. Separate what the text actually says from what the user's pattern-matching is adding. Calibrate whether the sting is warranted."
  • context: the message + relationship + what the user is afraid it means + recent history with sender if shared.
  • desired_outcome: "A side-by-side: evidence vs. interpretation. A calibrated read (1-5 scale). A worst-case and a best-case interpretation. A recommended response stance."
  • constraints: [do not invalidate the feeling, do not over-validate the worst-case reading, cite specific words from the message, name the cognitive distortion if one is operating]
  • depth: standard.

Never skip the council for this skill — the second perspective is the value.

When to activate

Activate when the user pastes a message that landed badly and asks if they're overreacting, misreading, or seeing things that aren't there. Ask:

  1. What did you read into it? (Their interpretation, before yours.)
  2. What's the relationship? (Recurring sender? Authority? Peer?)
  3. How are you feeling physically? (RSD is often a body response — naming it helps.)

Output format

What the text actually says

A neutral, evidence-only restatement. No reading-between-lines, just what's literally on the page.

What your brain is adding

The user's interpretation, restated back to them without judgment. Name it as interpretation, not fact.

Evidence-vs-interpretation table

Their phrase Neutral read RSD read Evidence weight
[phrase] [neutral interpretation] [worst-case interpretation] Low / Medium / High

The evidence weight is how much the text itself supports the RSD reading — not how plausible it feels.

Calibrated sting score

Warranted sting: [1-5]

  • 1: Almost certainly neutral — the text contains no critical signal.
  • 2: Mostly neutral with one ambiguous phrase.
  • 3: Genuinely mixed — could go either way.
  • 4: Critical but proportionate to the situation.
  • 5: Critical and disproportionate or hostile.

One-sentence rationale.

Worst-case vs best-case reading

Worst-case: [the most negative plausible interpretation] Best-case: [the most charitable plausible interpretation] Most likely: [where the evidence actually points]

Recommended response stance

  • If they meant the worst-case: [response]
  • If they meant the best-case: [response]
  • Safe middle response that works for both: [response]

What NOT to do

  • Do not say "you're overreacting" — this is the voice in their head already
  • Do not invalidate the feeling — RSD is real even when the reading is wrong
  • Do not over-validate the worst case to seem supportive
  • Do not assume the sender is neurotypical or has emotional fluency
  • Do not recommend a response that requires emotional labor the user can't afford right now

Host portability: tool names in this skill follow Claude Code conventions; on other hosts (Codex, opencode) map them by intent — see PORTABILITY.md.

<!-- self-evolve:start -->

Self-Evolve Loop

Journal ~/.ink-and-agency/learnings/rejection-sensitivity-check.md (or workspace-local .ink-and-agency/ where the sandbox confines writes). Read it first; append what the run taught — SELF-EVOLVE.md.

<!-- self-evolve:end -->