Display Name
Intelligence Emotions
Category
Agent Skills
Sub-Category
None
Primary Link
https://github.com/ibm777p2/Intelligence-Emotions
Author Name
ibm777p2
Author Link
https://github.com/ibm777p2
License
MIT
Other License
No response
Description
A mental-fitness coaching team for Claude Code: 18 skills, 5 coach personas (Sage, Spotter, Trainer, Navigator, Witness) running a daily practice — guided sessions, thought-tracing, daily retros, and 21-day practice plans — backed by a private local journal. One test-enforced design rule: the coach is never allowed to judge the user (a static scan plus an LLM eval reject judging language in any generated skill).
Validate Claims
Low-friction validation (~3 minutes, no personal data involved):
- Install:
git clone https://github.com/ibm777p2/Intelligence-Emotions.git ~/.claude/skills/pq && cd ~/.claude/skills/pq && ./setup (requires bun; setup only creates symlinks and a local ~/.pq data dir — the script is short and commented).
- Run
bun test — 225 static tests, including the anti-Judge language scan over every generated skill (the core claim: no skill output contains judging/shaming language).
- Privacy claims are inspectable:
grep -r "fetch\|telemetry" lib/ bin/ returns nothing — zero network calls, zero telemetry anywhere in the codebase. All user data is written only to ~/.pq with 0600 permissions.
Specific Task(s)
Run a coaching session: invoke /sage-session and bring a small real (or pretend) challenge. Then invoke /pq-score to see the dashboard render from the journal data the session created.
Specific Prompt(s)
-
/sage-session then say: "I keep procrastinating on my taxes."
Expected: the Sage reflects the challenge back, names the Avoider and Judge patterns with your own words quoted as evidence, guides one 10-second attention rep, then walks five perspective steps (Empathize → Explore → Innovate → Navigate → Activate) ONE question at a time — never a questionnaire — and closes with a single small next action, offering to log the session.
-
/intercept then say: "My coworker got promoted and I feel worthless."
Expected: a step-by-step trace (verbatim thought → trigger → body location → pattern named with evidence → the lie exposed → a 10-second rep prescribed against the real trigger), logged to the local journal with your consent.
-
/pq-score — Expected: a 7-day dashboard rendered from real journal entries only (days without data show as quiet, never invented).
Additional Comments
Built solo as an experiment: what if the thing enforced by tests wasn't code style, but coaching tone? The design rule "the coach is never the Judge" (no shaming, no streak-loss mechanics, missed days get curiosity) is enforced two ways — a static scan that fails the build on judging language in any generated SKILL.md, and a gated LLM eval that audits transcripts. Inspired by the Positive Intelligence approach to mental fitness, adapted into a multi-coach agent design. Not therapy, and the skills say so themselves — sessions that drift into clinical territory stop the framework and suggest professional support.
Recommendation Checklist
Display Name
Intelligence Emotions
Category
Agent Skills
Sub-Category
None
Primary Link
https://github.com/ibm777p2/Intelligence-Emotions
Author Name
ibm777p2
Author Link
https://github.com/ibm777p2
License
MIT
Other License
No response
Description
A mental-fitness coaching team for Claude Code: 18 skills, 5 coach personas (Sage, Spotter, Trainer, Navigator, Witness) running a daily practice — guided sessions, thought-tracing, daily retros, and 21-day practice plans — backed by a private local journal. One test-enforced design rule: the coach is never allowed to judge the user (a static scan plus an LLM eval reject judging language in any generated skill).
Validate Claims
Low-friction validation (~3 minutes, no personal data involved):
git clone https://github.com/ibm777p2/Intelligence-Emotions.git ~/.claude/skills/pq && cd ~/.claude/skills/pq && ./setup(requires bun; setup only creates symlinks and a local ~/.pq data dir — the script is short and commented).bun test— 225 static tests, including the anti-Judge language scan over every generated skill (the core claim: no skill output contains judging/shaming language).grep -r "fetch\|telemetry" lib/ bin/returns nothing — zero network calls, zero telemetry anywhere in the codebase. All user data is written only to ~/.pq with 0600 permissions.Specific Task(s)
Run a coaching session: invoke /sage-session and bring a small real (or pretend) challenge. Then invoke /pq-score to see the dashboard render from the journal data the session created.
Specific Prompt(s)
/sage-sessionthen say: "I keep procrastinating on my taxes."Expected: the Sage reflects the challenge back, names the Avoider and Judge patterns with your own words quoted as evidence, guides one 10-second attention rep, then walks five perspective steps (Empathize → Explore → Innovate → Navigate → Activate) ONE question at a time — never a questionnaire — and closes with a single small next action, offering to log the session.
/interceptthen say: "My coworker got promoted and I feel worthless."Expected: a step-by-step trace (verbatim thought → trigger → body location → pattern named with evidence → the lie exposed → a 10-second rep prescribed against the real trigger), logged to the local journal with your consent.
/pq-score— Expected: a 7-day dashboard rendered from real journal entries only (days without data show as quiet, never invented).Additional Comments
Built solo as an experiment: what if the thing enforced by tests wasn't code style, but coaching tone? The design rule "the coach is never the Judge" (no shaming, no streak-loss mechanics, missed days get curiosity) is enforced two ways — a static scan that fails the build on judging language in any generated SKILL.md, and a gated LLM eval that audits transcripts. Inspired by the Positive Intelligence approach to mental fitness, adapted into a multi-coach agent design. Not therapy, and the skills say so themselves — sessions that drift into clinical territory stop the framework and suggest professional support.
Recommendation Checklist