Brief Resume

brief-resume — load a prior brief and continue

You are loading a previously-written brief so this session can pick up where a prior session left off.

Step 1 — find the log

Compute the path yourself — no shell needed:

• ` = your current working directory with every / replaced by - (e.g. cwd /Users/you/code/app → slug -Users-you-code-app`). You already know the cwd from your environment.
• log file = /.claude/projects//briefs/brief-log.md (use your real absolute home path).

Read the log with the Read tool — reads are auto-allowed, so this never prompts for approval. If Read returns a not-found error, no briefs exist here: tell the user "No briefs found for this project. Have you run /brief-pause / /brief-delegate / /brief-report here before?" and stop.

Step 2 — show recent entries

From the log you just read, show the last 10 entries in reverse order (newest first), numbered 1-10. Example display:

Recent briefs (newest first):

 1. 2026-05-23 03:15 · pause    · Export feature-complete; empty-response edge case open
 2. 2026-05-23 02:55 · report   · Parser done; unknown-codec case accepted
 3. 2026-05-23 02:10 · delegate · Wire up the fast-path in the probe module
 4. 2026-05-22 19:45 · pause    · Cache rework paused mid-stream
 ...

Which one would you like to resume? (number, or "latest" for #1)

Step 3 — load the chosen brief

Read the brief file the user picked (or #1 if they said "latest" / nothing) with the Read tool. Resolve the path as /.claude/projects//briefs/ from the log entry.

If the file no longer exists, it was pruned (only the ~60 most recent brief files are kept; the log holds up to 50 entries, so this is rare and only hits very old picks). Tell the user the file was pruned, show its one-line log summary as the only remaining record, and ask them to pick a different entry.

Step 4 — acknowledge + continue

After reading, do all three:

1. Tell the user briefly what you loaded (the title + flavor).
2. Read any docs the brief's Read first: line points to — but only if they're project-local (CLAUDE.md, docs/**). Don't chase external links.
3. Take the action implied by the brief's flavor:

• pause → "Picking up from where we left off. Doing the first item under Do next unless you redirect."
• delegate → "Executing the delegated task per Steps. I'll report back when done via /brief-report."
• report → "Loaded the report-back. Reviewing What changed and What didn't / blockers before continuing."

Do NOT auto-execute commits, deploys, or anything destructive — confirm with the user first even if the brief said to do them.

What NOT to do

• Don't read briefs older than #10 unless the user explicitly asks ("show me older briefs", "resume from May 22").
• Don't read multiple briefs at once — that defeats the point of the index.
• Don't read the brief and immediately start writing code — confirm the action with the user first.
• Don't auto-trigger /brief-pause at the end of resumed work — the user invokes that explicitly when ready.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: useaitechdad
• Source: useaitechdad/agent-skills
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.