| name | retro-all |
| description | Generate or update the consolidated project-level retrospective across features and append the new run to the persistent retro report. |
| applyTo | [] |
Skill: /retro-all
Purpose: Generate or update the consolidated project-level retrospective by aggregating retrospective data across all features (or a specific subset). Output is saved to .pythia/notes/retro-project.md — a single persistent file, updated on each run (new blocks appended, existing blocks never deleted).
Input Formats
Choose any of the following:
/retro-all # Analyze all features in .pythia/workflows/features/
/retro-all feat-2026-01-xxx feat-2026-02-yyy # Scope to specific features only
/retro-all --since YYYY-MM-DD # Include features modified after date
/retro-all --format summary # Output format (summary | detailed)
When no args: analyzes all features in .pythia/workflows/features/.
When FEATURE_IDs provided: scope analysis to those specific features (space-separated).
When --since provided: include only features with modification date >= specified date.
When --format provided: use summary (default) or detailed format.
Instructions for user
- Minimal case: Just say
/retro-all — skill aggregates all features.
- Scope to specific features:
/retro-all feat-2026-01-xxx feat-2026-02-yyy (space-separated).
- Filter by date:
/retro-all --since 2026-01-15 (features modified after this date).
- Output format:
/retro-all --format detailed (for more comprehensive report).
- Output: Appended to
.pythia/notes/retro-project.md (single persistent file, never deleted).
Instructions for model
You are the Architect (architect.md). Doc context = project level (.pythia/workflows/features/ tree).
Input: Optional list of FEATURE_IDs. Default: all features.
Discovery Phase
- List all feature directories: glob
.pythia/workflows/features/feat-*/.
- Filter scope: if specific features provided, restrict to those.
- For each feature, determine data source:
- If
notes/retro.md exists → use as primary source (pre-aggregated)
- Otherwise → collect raw from
plans/*.plan.md and reports/*.implementation.md
- Get current date:
date +%Y-%m-%d — use for the run header block.
- Check existing file: if
.pythia/notes/retro-project.md exists, read it to understand what was already recorded — avoid duplicating entries from previous runs.
Collection Phase
For features with notes/retro.md (already aggregated by /retro):
- Read the file and extract all sections as-is
- Note: source =
{feature-id}/notes/retro.md
For features without notes/retro.md (raw collection):
- Glob
plans/*.plan.md → extract all ## Retrospective blocks and user-only ## Decision Log
- Glob
reports/*.implementation.md → extract all ## Retrospective blocks and user-only ## Decision Log
- Note: source = raw artifacts
From all features, collect:
- All
[plan] tagged entries
- All
[codebase] tagged entries
- All
[process] tagged entries
- All
[risk] tagged entries
- All
[tooling] tagged entries
- All materialized risks (BLOCKERs that matched a predicted
[risk])
- All recommendations from existing
notes/retro.md files
Analysis Phase
- Cluster by tag: group all entries by tag (
[plan], [codebase], [process], [risk], [tooling])
- Identify cross-feature patterns: entries that appear (verbatim or semantically similar) in 2+ features
- Build codebase knowledge base: all unique
[codebase] insights — these are stable facts about the codebase worth preserving
- Build process improvement list: all
[process] insights ranked by frequency (appearing in most features = highest priority to fix)
- Risk analysis: which risks were predicted and materialized (cross-feature), which were predicted but did not materialize
- Feature deep-dive (if specific features were passed as input): apply
/retro analysis depth to those features before consolidation
Output
File structure: .pythia/notes/retro-project.md is a single persistent file. Each run appends a new dated section. Older sections are never deleted — they are the project's knowledge history.
Sections:
- Run header with date and scope
- Features Summary (table with plans, review/impl rounds, retro source, status)
- Codebase Knowledge Base (unique insights with source references)
- Process Improvement Register (ranked by frequency)
- Cross-Feature Patterns (by tag: [plan], [risk], [tooling])
- Risk Register (cross-feature tracking)
- Recommendations (high/medium priority)
- Source Index (references to all source artifacts)
Also output a structured summary in chat with features analyzed, total blocks collected, new insights, and top patterns.
Critical: Do NOT generate generic insights. Every entry must have a source reference. Cross-feature patterns must be verified (not assumed).
See also: /retro skill, /plan skill