A period report for a reader outside the session — a weekly update, client status, "what did we get done this month". Skip for end-of-task reports (superstack-ship), retros when a retro tool ships, and recalling past sessions for yourself (a memory tool).
Scanned 9/19/2026
Install to Claude Code
npx -y skills add debabsah/superstack --skill superstack-digest --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Superstack Digest?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/debabsah-superstack-digest)More formats (shields.io, HTML) on the badges page.
---
name: superstack-digest
description: A period report for a reader outside the session — a weekly update, client status, "what did we get done this month". Skip for end-of-task reports (superstack-ship), retros when a retro tool ships, and recalling past sessions for yourself (a memory tool).
---
# superstack-digest
The digest is the calibration record paying rent outward. It is assembled from records, never from memory — the conversation is scratch; files are the project (superstack-continuity's rule, applied to reporting).
The description's deferrals (a retro tool for retrospectives; a memory/session-history tool for self-recall) bind only if the named tool is really present — check the session's skill listing first and announce the hand-off; when nothing is there, assemble from records right here.
## Assemble
1. **Gather the period's facts from the record:** `git log` across the repos in scope, the claims-log's `Verified:` lines, value-log settlements (`HELD`/`MISSED`), queue movements, open residuals and skipped-gates. Your recall of the period is a hypothesis; the logs are the evidence (R1 on your own memory). **If the record is thin** — early days, files not yet grown — say so: build from `git log` alone and ledger the gap (`Assumed: the period wasn't fully logged — built from git history only`).
2. **Numbers in a digest are data claims.** Sanity-check each before quoting — counts from commands, reconciled against a known total where one exists. A `PROVISIONAL` number never ships to an external reader: round it into honest words or leave it out.
## Write for the named reader
3. **Their language, not the method's.** Shipped work in outcome words, not commit words ("the report now runs in under a minute", not "refactored the aggregation pipeline"). No session jargon, no tier or ledger tokens **in the artifact** — the draft the client reads goes tokenless. **The chat turn that offers the draft is a normal report and carries its ledger** (`Verified:` what you assembled and from which logs, `Assumed:` what the record couldn't show) — the claims gate reads your turn, not the attachment, and the grammar leaves the artifact, never the method. The honesty travels into the artifact anyway: "on track" is a claim; write it only when the record supports it, and never soften a real problem (R5 holds even in business register — *especially* there).
4. **The reader's sections:** done · in progress with expected next · **blocked / asks** (what you need from *them*, explicit — a digest that hides its asks is decorative) · risks worth their attention.
5. **End the same way every time:** where things stand, what happens next and whose move it is.
## The boundary
**Draft, don't send.** Write the draft to a file named for the period (`<period>-digest.md`, or where the human asks) — outward-facing, so sending is the human's act (the inward/outward line). Offer it with anything you'd flag before it travels — a number you couldn't verify, a claim you'd want them to gut-check — stated above the fold, not discovered by the client.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!