CtrlK
BlogDocsLog inGet started
Tessl Logo

pantheon-ai/retrospect-collab

Analyse human-AI collaboration patterns and compute quality metrics from captured session data.

88

Quality

88%

Does it follow best practices?

Impact

Pending

No eval scenarios have been run

SecuritybySnyk

Risky

Do not use without reviewing

Overview
Quality
Evals
Security
Files

reference.md

retrospect-collab Reference

Report Template

# Collaboration Retrospective: YYYY-MM-DD to YYYY-MM-DD

## Sessions Analyzed
- [List with prompts/tools counts]

## Part 1: Technical Effectiveness

### Context Management
**What worked well:** [Specific examples]
**What didn't work:** [Specific examples]
**Actions:** - [ ] [Specific improvements]

### Guidance & Direction
**Clarity wins:** [Effective prompt examples]
**What didn't work:** [Vague/problematic prompt examples]
**Actions:** - [ ] [Specific improvements]

### Critical Thinking
**Effective challenges:** [Productive pushback examples]
**Missed challenges:** [Where questioning would have helped]
**AI challenged me:** [Helpful AI pushback examples]
**Actions:** - [ ] [Specific improvements]

### Bias & Blindspots
**Biases noticed:** [Specific examples]
**Blindspots AI helped overcome:** [Examples]
**Blindspots AI introduced:** [Examples]
**Actions:** - [ ] [Specific improvements]

---

## Part 2: Cognitive Posture & Professional Development

### Intentionality & Learning Posture
**What worked well:** [Intentional design, concept understanding examples]
**What didn't work:** [Copy-paste, trial-and-error examples]
**Actions:** - [ ] [Specific improvements]

### Agency & Ownership
**Evidence of agency:** [Decision ownership, custom prompts, self-correction]
**Passive patterns:** [Template following, AI dependency]
**Actions:** - [ ] [Specific improvements]

### Impact Categorization

**Breakdown:**
- **High-impact** (Z%): [Key sessions with examples]
- **Low-impact** (Y%): [Sessions with patterns]
- **Automation** (X%): [Sessions with patterns]

**Triggers for high-impact:** [What preparation/framing led to transformative sessions?]
**Regression triggers:** [What caused drops? Time pressure? Unclear goals?]

### Skill Progression (Longitudinal)
**Evidence of growth:** [Session 1→N comparisons]
**Stagnation patterns:** [Repeating mistakes]
**Prompt evolution:** [Simple → Structured → Advanced]

---

## Start/Stop/Continue

### Start
**Technical:** [New collaboration practices]
**Cognitive:** [New learning/agency practices]

### Stop
**Technical:** [Anti-patterns]
**Cognitive:** [Dependency patterns]

### Continue
**Technical:** [Effective patterns]
**Cognitive:** [Effective learning patterns]

---

## Metrics Summary
- **Sessions**: N | **Total prompts**: X (avg Y) | **Total tool calls**: X (avg Y)
- **Total duration**: Xh Ym (avg Ys/session) | **Subagents**: X
- **Impact**: High Z% (>60% target) | Low Y% (20-30%) | Auto X% (<20%)

## Action Items
**Priority 1 (Technical):** [Specific improvements]
**Priority 2 (Cognitive):** [Specific improvements]

Technical Effectiveness Analysis Questions

Context Management:

  • How was context prepared before sessions?
  • Evidence of context over-dump (many reads without using info)?
  • Evidence of context under-dump (repeated clarification questions)?
  • Use of CLAUDE.md, specs, structured context?
  • Progressive feeding vs all-at-once?

Guidance & Direction:

  • Clarity of initial prompts (specific vs vague)
  • Balance of exploration vs constraints
  • Over-constraining vs under-constraining patterns

Critical Thinking:

  • When did user challenge AI outputs?
  • When should user have challenged but didn't?
  • When did AI challenge user assumptions?
  • Pattern of accepting first solution vs requesting alternatives

Bias & Blindspots:

  • Over-trust: accepting without verification
  • Dismissal: ignoring helpful suggestions
  • Confirmation bias: only accepting matching outputs
  • Automation bias: preferring AI solutions over manual review

Impact Indicators

Automation: Repetitive tasks, simple summaries, copy-paste prompts, could be a script

Low-impact augmentation: Basic Q&A, incremental improvements, standard templates, surface-level

High-impact augmentation: Transformative insights, complex problem-solving, custom prompts, professional skill amplification, novel approaches, 6-block structured prompts, capability-building

Cognitive Posture Analysis

Intentionality signals: Structured prompts, clear objectives, "why before how", concept understanding, reusable patterns

Agency signals: Custom prompt construction, template adaptation, decision ownership, self-correction, tool mastery progression

Progression tracking: Prompt complexity evolution, concept mastery vs memorization, strategic vs tactical usage, sustainable growth vs beginner's luck

reference.md

SKILL.md

tile.json