Skip to content

mostafa-drz/claude-skills / code/emotional-recap

emotional-recap

Verified

Reviews the user's past Claude Code conversations from a wellbeing perspective — sentiment, tone, emotional arc, recurring patterns — and generates a supportive, science-grounded report in both Markdown and HTML. Default lookback is 48 hours across all projects. Uses recognised emotion frameworks (Plutchik, Ekman, Russell's circumplex, Pennebaker linguistic markers) and cites the science behind every observation. Learns the user's baseline tone over time so future reports flag genuine shifts, not noise. Use when the user asks for an emotional/wellbeing recap, mood check, sentiment review, or wants to understand their own ups and downs across recent work sessions.

FileKindSize
DESIGN.mdmarkdown42.7 KB
SKILL.mdskill md25.7 KB
icon.svgother481 B
reference/citations.mdmarkdown7.6 KB
reference/frameworks.mdmarkdown7.1 KB
reference/report-structure.mdmarkdown36.1 KB

reference/frameworks.md

View on GitHub
# Emotion frameworks reference

This file is loaded by the main workflow when applying sentiment / emotion analysis. Each framework section gives the model the exact taxonomy and decision rules to use, plus a short justification.

---

## 1. Plutchik's Wheel of Emotions (1980)

**Eight primary emotions**, arranged in opposing pairs:

| Emotion       | Opposite      | Linguistic markers (English)                              |
|---------------|---------------|-----------------------------------------------------------|
| Joy           | Sadness       | "great", "love", "amazing", "yes!", excited punctuation   |
| Trust         | Disgust       | "thanks", "appreciate", "agree", reassurance language     |
| Fear          | Anger         | "worried", "afraid", "what if", hedging modals            |
| Surprise      | Anticipation  | "wait", "didn't expect", "huh", exclamatives              |
| Sadness       | Joy           | "tired", "down", "ugh", flat affect, low energy markers   |
| Disgust       | Trust         | "gross", "awful", "hate", strong rejection language       |
| Anger         | Fear          | "frustrated", "stupid", "wtf", caps, repeated punctuation |
| Anticipation  | Surprise      | "looking forward", "planning", future-tense + agency      |

**Intensity gradients** (Plutchik called these dyads): each primary has a milder and stronger form, e.g. _serenity → joy → ecstasy_. The skill records the primary; intensity comes from the Russell arousal axis (below).

**Decision rule**: tag a user message with **0-2 primary emotions**. If neither tag has strong evidence, leave empty (tone-neutral). Never assign all 8 — that's noise.

**Why this framework**: Plutchik's wheel is the most widely-used emotion model in computational sentiment work because it gives discrete categories that pair naturally with linguistic markers. It's used in NRC Emotion Lexicon (Mohammad & Turney, 2013), which is the de facto standard in NLP emotion tagging.

---

## 2. Ekman's Basic Emotions (1992)

Six universal emotions argued to be cross-culturally recognised:

- Happiness
- Sadness
- Fear
- Anger
- Surprise
- Disgust

Ekman's set overlaps Plutchik's but drops Trust and Anticipation. Useful when:
- The conversation is short and you want a coarser tag
- The user preference is `framework: ekman`
- You're producing the "dominant emotion" line in the at-a-glance section — six is easier for humans to parse than eight.

**Why this framework**: Ekman's facial-expression studies (1971, 1992) are the foundational empirical case for universal emotion categories. Even with later challenges (Barrett, 2017, _How Emotions Are Made_), Ekman remains the canonical reference and is widely understood.

---

## 3. Russell's Circumplex Model of Affect (1980)

A **dimensional** rather than categorical model. Every affective state is a point on two axes:

- **Valence** — pleasant (+1) ↔ unpleasant (−1)
- **Arousal** — activated (+1) ↔ calm (0)

This produces four quadrants:

| Quadrant | Valence | Arousal | Example states           | Suggested HTML colour swatch |
|----------|---------|---------|--------------------------|------------------------------|
| Q1       | +       | +       | excited, enthusiastic    | warm gold `#d4a04a`          |
| Q2       | −       | +       | tense, frustrated, angry | desaturated rose `#c66b6b`   |
| Q3       | −       | −       | tired, bored, low        | dusty indigo `#6b6b8a`       |
| Q4       | +       | −       | calm, content, focused   | sage green `#7a9580`         |

**Decision rule**: every analysed user message gets a `(valence, arousal)` pair in `[-1, 1] × [0, 1]`. Aggregate by mean.

The **dominant quadrant** for the report is the quadrant where the centroid of all message points falls. Use this to pick the at-a-glance swatch colour.

**Why this framework**: Russell's circumplex is the dominant dimensional model in affective science. It explains why "tired" and "frustrated" — both negative — feel completely different (arousal axis), and why categorical tags miss that. The valence/arousal pair is also what most affective-computing systems output today (Russell, 1980; Posner, Russell, & Peterson, 2005).

---

## 4. Pennebaker-style linguistic markers (LIWC, 2003–2015)

The skill applies a lightweight LIWC-inspired analysis. The model counts (or estimates ratios for) these categories across the window:

| Category              | What it suggests                                              | Reference                        |
|-----------------------|---------------------------------------------------------------|----------------------------------|
| 1st-person singular   | Self-focus; rising trend correlates with rumination / sadness | Pennebaker (2003), Rude et al. (2004) |
| Negative-emotion words| Distress, frustration, anxiety                                | Tausczik & Pennebaker (2010)     |
| Positive-emotion words| Engagement, gratitude, satisfaction                           | Tausczik & Pennebaker (2010)     |
| Cognitive complexity  | "because", "however", "although" — flexible thinking          | Pennebaker (2003)                |
| Certainty markers     | "never", "always", "must" — rigid thinking, often anxious     | Pennebaker (2003)                |
| Social references     | "we", "they", names — connection vs isolation                 | Tausczik & Pennebaker (2010)     |
| Temporal focus        | Past / present / future verb tense ratio                      | Pennebaker (2011)                |

**Decision rule for flagging**: only flag a marker shift if it differs **≥50% from baseline** (learned from prior sessions). Never flag on the first session — there's no baseline yet.

**Important caveat to include in reports**: linguistic markers are correlational, not diagnostic. A high 1st-person count in a debugging session doesn't mean depression — it might mean the user is the only one debugging. Context matters.

---

## 5. Supporting frameworks (cited but not used as primary)

- **Csikszentmihalyi's Flow (1990)** — productive struggle vs. distress. Used to reframe frustration patterns that show high engagement (long messages, complex problem-solving) rather than withdrawal.
- **Fredrickson's Broaden-and-Build (2001)** — positive emotions broaden cognition and build resources. Used in the "Bright spots" section to explain _why_ that section matters.
- **Walker's sleep research (_Why We Sleep_, 2017)** — invoked when late-night sessions correlate with negative-valence drops.
- **Barrett's Theory of Constructed Emotion (2017)** — invoked in the caveats section to remind the reader that emotion categories themselves are constructed, not discovered. This is the honest counterweight to Ekman.

---

## Decision priority

When `framework: combined` (the default):

1. Use **Russell** for the at-a-glance quadrant + sparkline (it's the most informative single number).
2. Use **Plutchik** for the "Dominant emotions" section (richer categorical labels).
3. Use **Pennebaker** for the "Patterns" section (linguistic shifts over time).
4. Use **Ekman** as a fallback simplification when summarising for the user verbally.

When `framework: <single>`, use only that one across all sections.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall