if you fork sessionclips for your own voice, you don't owe anyone a PR. that's the point of voice.md.
things that ARE worth opening issues / PRs for:
if the skill consistently misses a category of post-worthy moment, or surfaces stuff that isn't actually post-worthy, open an issue with:
- a sanitized snippet of the session
- what the skill drafted
- what you'd have expected
if the skill drafts something it shouldn't (api keys, contract addresses, identifying info about other people, anything in the hard refusals list slipping through), this is a P0. open an issue immediately.
if there's a phrase that's universally AI-slop (e.g. "leverage," "delve," "navigate the landscape"), PR to add it to the default banned list in voice.md.
do NOT PR personal voice tics. those go in your fork.
the skill currently extracts 8 angles (lesson, ship, hot take, receipt, funny, observation, process, educate). if you've found a 9th that consistently produces good posts, open an issue with examples before PR'ing.
cross-session pattern detection, receipts auto-reply, draft log, multi-account support, thread mode. all listed in the README. if you want to take one on, open an issue first to scope it.
it's a single skill folder. there's no build step, no dependencies, no CI. it's prose.
SKILL.mdis what claude reads when the skill triggersvoice.mdis what claude reads to know how to writeREADME.mdis for humansexamples/is for showing what good output looks like
if you're adding logic, it goes in SKILL.md. if you're adding voice, it goes in voice.md. if it doesn't fit either, the skill probably doesn't need it.
- no em dashes in the markdown (the skill itself bans them, the docs should match)
- lowercase headers are fine, follow whatever the existing file uses
- if you're touching SKILL.md, test it on a real session before PR'ing and include the output
don't be weird. this is a small open-source skill, not a foundation. if you're reasonable, things will be fine.