Back to Skills

weekly-digests

Generate a serial week-by-week narrative digest of a project's full claude-mem timeline. Splits the timeline into per-ISO-week files, then runs one consecutive subagent per week — each receiving the prior week's carry-forward block — to produce one chapter per ISO week of data. Use when asked for "weekly digests", "week-by-week story", "serial timeline", or "narrative chapters" of a project's history.

86,386stars7,474forksUpdated 7/8/2026

Security Assessment

Safe(100/100)
Security Score100/100

About weekly-digests

Generates a serial, multi-chapter narrative digest of a project's complete claude-mem history, producing one chapter per ISO week of data. It differs from a single sweeping timeline-report: the timeline is split into per-ISO-week files, and one consecutive subagent runs per week, each receiving the prior week's carry-forward block so the story stays coherent across chapters. The chapter count is not fixed; it equals the number of ISO weeks the timeline covers, so a two-week project yields two chapters and a thirty-week project yields thirty.

The workflow resolves the claude-mem worker port, determines the project name (using the parent project when run inside a git worktree), and fetches the full timeline from the worker's context-inject endpoint into a scratch file. A Python splitter parses date headers, groups days into ISO weeks via date.isocalendar() (Monday-start), and writes one file per week under docs/timeline-weeks/, preserving each day's section verbatim and running a dual-pass sanity check that total observations are conserved. A README index table (Week, Dates, Observations, Sessions, File) is built as an operator roadmap, and the resulting file count becomes TOTAL, which drives the pipeline; empty weeks between active weeks are skipped.

Use it when someone asks for weekly digests, a week-by-week story, a serial timeline, or narrative chapters of a project's history, as opposed to a single report. The heart is a strictly sequential subagent pipeline: for each week in chronological order, a general-purpose subagent is dispatched with a prompt template that includes the source week file, an output digest path, a format key for the observation and session lines and emoji legend, and the prior week's carry-forward. Each digest includes a title, tagline, narrative, threads continued/opened/resolved, a cliffhanger, and a fenced carry-forward block capped at about 350 words with disciplined pruning. Tone rules ask for a sharp third-person narrator that treats recurring codebase components as characters, tracks the user's prompt-register evolution, and, for quiet weeks, treats the silence itself as the story.

FAQ

How many chapters does a digest produce?

One chapter per ISO week the timeline covers; the count is not fixed, so two weeks of data give two chapters and thirty weeks give thirty.

How does this differ from timeline-report?

timeline-report produces one long report, whereas weekly-digests produces a serial chapter per ISO week, with each subagent reading the prior week's carry-forward so the narrative stays coherent.

Why must the subagents run sequentially rather than in parallel?

Each agent receives the previous agent's carry-forward block; running them in order is what turns N disjoint summaries into a single N-chapter serial narrative.

What are the prerequisites?

A running claude-mem worker, a project with at least one ISO week of observations (even N=1 works), and a clean output directory to write into.

What must each week's carry-forward block contain?

It is fenced and capped at about 350 words, aggressively pruned, with Active arcs, Cast, Unresolved, and Tone notes sub-sections for the next week's agent.

Install weekly-digests

Download and extract the skill files to your .claude/skills/ directory.

Quick Setup:

  1. Copy the skill folder to .claude/skills/
  2. Claude will automatically detect and use the skill