What
KNOWN_ISSUES.md notes 'No baselines ship in-repo.' First-time users running the README's 5-min quick start get verdict: FAIL (no baseline diff). This is correct behavior but a bad first impression.
Design
Two parts:
Part 1: A scripted demo
Add scripts/demo.sh that does the whole tutorial end-to-end:
#!/usr/bin/env bash
# Walks a new user through baseline → edit → detect-regression
set -e
echo "== Step 1: baseline omo-session-distiller (uses your real ANTHROPIC_API_KEY) =="
eval-harness baseline --skill=omo-session-distiller
echo
echo "== Step 2: mutate the skill (drop a tag) =="
# Edit a known line
sed -i.bak 's/architecture,database/architecture/' skills/omo-session-distiller/SKILL.md # example
echo
echo "== Step 3: re-run; harness should detect regression =="
eval-harness run --skill=omo-session-distiller
echo
echo "== Step 4: restore =="
mv skills/omo-session-distiller/SKILL.md.bak skills/omo-session-distiller/SKILL.md
Part 2: a recorded asciinema cast under docs/demo.cast (or animated gif) that a contributor can replay without their own API key.
Scope
demo.sh itself is small (~30 lines). The asciinema cast is nice-to-have but slows down review. Start with just the script + a brief README pointer.
What
KNOWN_ISSUES.mdnotes 'No baselines ship in-repo.' First-time users running the README's 5-min quick start getverdict: FAIL(no baseline diff). This is correct behavior but a bad first impression.Design
Two parts:
Part 1: A scripted demo
Add
scripts/demo.shthat does the whole tutorial end-to-end:Part 2: a recorded asciinema cast under
docs/demo.cast(or animated gif) that a contributor can replay without their own API key.Scope
demo.shitself is small (~30 lines). The asciinema cast is nice-to-have but slows down review. Start with just the script + a brief README pointer.