python

skillscope (Python POC)

Original Python proof-of-concept for skillscope. Superseded by the Rust implementation at the repo root — kept here as the parity oracle ../../scripts/parity.sh (make parity from the repo root) checks the Rust export output against, and as a reference for the core parsing/fidelity logic.

Not the shipped tool. See the root README for current status.

Run

uv run skillscope summary # per-skill counts + trigger breakdown
uv run skillscope sessions <skill># session drill-down for one skill
uv run skillscope timeline [skill] # time-series (daily/weekly)
uv run skillscope projects # per-project breakdown
uv run skillscope fidelity # trigger-fidelity report
uv run skillscope export# JSON export of normalized invocations

Known limitations (frozen as of the bake-off)

• Fidelity heuristic is deliberately loose (>=2 keyword hits = plausible match); the Rust implementation replaced this with TF-IDF weighting.
• Subagent transcripts (<session>/subagents/agent-*.jsonl) are excluded — skill invocations made by subagents are not counted.
• sessions-index.json is not joined for friendlier session labels.

Architecture

src/skillscope/
├── models.py # SkillInvocation, SkillDefinition, UserPrompt, TriggerType
├── parser.py # JSONL streaming extraction → SkillInvocation / UserPrompt
├── aggregate.py # counts, trigger breakdown, time-series, per-project rollups
├── fidelity.py # skill discovery + trigger-fidelity heuristics
└── cli.py # argparse subcommands, rich tables, JSON export