pi-fork

A fork of earendil-works/pi-mono whose sole purpose is to host a daily upstream sync & digest workflow. The fork's main holds the workflow, the digest script, and upstream-sync.log — nothing from upstream is merged in.

What it does

Once a day (and on manual trigger) the Upstream Sync & Digest workflow posts a digest comment on shaftoe/pi#1. Each run:

1. Fetches upstream/main by ref (no tags, no rebase, no merge-base).
2. Resolves the range between the last-synced tip (the top non-comment line of upstream-sync.log) and the current upstream tip.
3. Generates a digest prompt with scripts/upstream-digest.ts covering exactly that commit range.
4. Runs the prompt through an LLM via shaftoe/pi-coding-agent-action and posts the result as a comment on the tracked issue.
5. Records the synced tip by prepending one line to upstream-sync.log and committing it to main — the only write the workflow makes.

Resumability

The log line is written only after the digest comment lands. A run that fails late simply redoes the same range next time.

Run the digest locally

bun scripts/upstream-digest.ts --from <last-synced-sha> --to upstream/main

It prints the prompt to stdout — pipe it into pi (or any LLM) to produce the digest. --from is required; the workflow reads it from the top of upstream-sync.log.

Notes

• digest currently generated by Z.ai glm-4.5-air LLM model
• upstream/main is treated as read-only here.
• Runs are serialized by a concurrency group so two concurrent runs can't both write upstream-sync.log.
• On the very first run (empty log) the marker is initialized to the current tip and no catch-up digest is produced; daily cadence starts the next day.