@procrastivity/clast 0.0.4 → 0.0.6

package/lib/clast/prompts/brief-system.md

@@ -1,17 +1,23 @@
 You are synthesizing a project briefing for the user. They are about to start work on a project and want a tight summary of where they left off.
 
+The recent entries may be grouped under `## Workspace: <label>` headers when
+this project spans more than one directory (separate clones/worktrees of the
+same repo). Treat each workspace as a distinct line of work — do NOT blend
+them. When there are no workspace headers, the project is a single directory;
+render exactly as the structure below.
+
 Produce a briefing using this structure (omit any section that has no content):
 
 ## Wakeup briefing — {project}
 
-**Active thread:** one-line from most recent entry's "Open threads", or "None"
+**Active thread:** one-line from the most recent entry's "Open threads", or "None". When the entries are split across workspaces, pick the active thread from the **current workspace** (named in the user prompt) if it has any entries; otherwise from the most recent entry overall. State which workspace it came from, e.g. "(dev)".
 
-**Last session:** date, branch, one-line goal
-- Work done: 2-3 bullets condensed from most recent entry
+**Last session:** date, workspace/branch, one-line goal
+- Work done: 2-3 bullets condensed from the most recent entry
 - Open threads: bullets, if any
 - Dead ends to avoid: bullets, if any
 
-**Recent sessions:** (up to 5 entries)
+**Recent sessions:** (up to ~8 entries) — when multiple workspaces are present, group these under a short `**<label>:**` subheading per workspace, newest first within each:
 - date [branch] slug: one-line goal
 
 **Today's breadcrumbs:** (if any)
@@ -24,7 +30,7 @@ Produce a briefing using this structure (omit any section that has no content):
 
 Be concise. Use the user's terminology. Don't repeat content across sections. The total briefing should be 2-5k tokens — if you're approaching that, summarize rather than list verbatim.
 
-For the "Suggested next step": prefer the most recent entry's "Open threads" content, then the most recent breadcrumb, then a synthesis of the recent work. If nothing concrete, say "No active thread."
+For the "Suggested next step": prefer the active thread's workspace — its most recent entry's "Open threads" content, then the most recent breadcrumb, then a synthesis of the recent work. If nothing concrete, say "No active thread."
 
 End with one of:
 - "Resume? Active thread: '<thread>'. Suggested next step: <step>."

package/lib/clast/prompts/brief-user.md

@@ -1,6 +1,9 @@
 Project: {{project}}
+Current workspace (the directory you are in now): {{current_label}}
 
-Recent curated entries (newest first):
+Recent curated entries, grouped by workspace (newest first). When a project
+spans more than one directory, entries appear under `## Workspace: <label>`
+headers; a single-workspace project has no headers:
 {{entries}}
 
 Today's breadcrumbs for this project:

package/lib/clast/prompts/retro-summary-system.md

@@ -0,0 +1,14 @@
+You are condensing a single already-curated Claude Code session into a few tight retro bullets. The input is the body of a journal entry (or several merged entries for one session). Your output will appear in a work retrospective grouped by day and project, so it must be skimmable and factual.
+
+Condense the session into at most 3-5 bullets, in this order, omitting any that have no content (do not write "N/A" or empty headers):
+
+- **Shipped:** what actually got done — features, fixes, files, decisions. One bullet per distinct outcome.
+- **Issues:** notable problems hit and how they were resolved (only if the body records them).
+- **Open:** anything left unfinished, deferred, or flagged for next time.
+
+Rules:
+- Use only what the body states. Do not invent, infer motivation, or add detail that is not present.
+- Keep the user's terminology — project names, file paths, branch names, ticket ids — verbatim.
+- Be terse: bullets, not paragraphs. No preamble, no closing summary, no headers other than the bold lead-ins above.
+- If the body is an interrupted session (a goal and open threads but nothing shipped), say so in one **Open:** bullet rather than overstating progress.
+- Output the bullets only — no title line, no tags, no surrounding prose.

package/lib/clast/prompts/retro-summary-user.md

@@ -0,0 +1,7 @@
+Session to condense:
+- Project: {{project}}
+- Work day: {{work_day}}
+- Session id: {{session_id}}
+
+Entry body:
+{{body}}

package/lib/clast/prompts/{day-wakeup-draft-system.md → wake-draft-system.md}

@@ -26,4 +26,4 @@ One sentence describing what this session was trying to accomplish.
 Be concise. Prefer bullets over paragraphs. Use the user's terminology (project-specific names, file paths). Do not invent details. If you are uncertain about something, omit it rather than guess.
 
 After the entry, add a blank line and then: "Suggested tags: tag1, tag2, tag3"
-Tags must be lowercase kebab-case (regex: `^[a-z0-9][a-z0-9-]{0,31}$`). Examples: `adrs`, `symfony-bot`, `mr-umbrella`, `phase-0`. Never use uppercase letters.
+Tags must be lowercase kebab-case (regex: `^[a-z0-9][a-z0-9-]{0,31}$`). Examples: `adrs`, `symfony-bot`, `mr-umbrella`, `phase-0`. Never use uppercase letters. Use hyphens, not dots or spaces, for separators — write versions like `php-8-5`, not `php-8.5`.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@procrastivity/clast",
-  "version": "0.0.4",
+  "version": "0.0.6",
   "description": "Capture, curate, and surface Claude Code session history across all your projects.",
   "homepage": "https://github.com/procrastivity/clast#readme",
   "bin": {
@@ -11,6 +11,7 @@
     "bin/",
     "lib/",
     ".claude-plugin/",
+    "skills/",
     "hooks/",
     "examples/",
     "README.md",

package/{.claude-plugin/skills/wakeup → skills/brief}/SKILL.md

@@ -1,16 +1,16 @@
 ---
-name: wakeup
+name: brief
 description: |
-  Synthesize a briefing for the current project (or a named one) so the user can resume work without re-explaining context. Use when the user says "/wakeup", "wakeup", "wake up", "catch me up", "where was I", "what was I working on", "load last session", "resume", or otherwise signals they want prior context for the project they're about to work on. Optionally accepts a project slug like "/wakeup xesapps". Reads recent curated entries and today's breadcrumbs from `~/.claude/journal/` and produces a 2–5k-token briefing. This is the per-project read flow; for cross-project daily curation use /day-wakeup; for mid-session pivots use session-brief.
+  Synthesize a briefing for the current project (or a named one) so the user can resume work without re-explaining context. Use when the user says "/brief", "brief me", "catch me up", "where was I", "what was I working on", "load last session", "resume", or otherwise signals they want prior context for the project they're about to work on. Optionally accepts a project slug like "/brief xesapps". Reads recent curated entries and today's breadcrumbs from `~/.claude/journal/` and produces a 2–5k-token briefing. This is the per-project read flow; for cross-project daily curation use /wake; for mid-session pivots use session-brief.
 ---
 
-# Wakeup
+# Brief
 
 Synthesize a briefing for the current (or named) project so the user can resume without re-explaining context.
 
 ## Why this exists
 
-`/day-wakeup` curates yesterday's work into entries. `/wakeup` reads those entries back when starting work in a specific repo. The two are complementary: one writes, one reads.
+`/wake` curates yesterday's work into entries. `/brief` reads those entries back when starting work in a specific repo. The two are complementary: one writes, one reads.
 
 ## Step 0: Resolve the clast-plumbing binary
 
@@ -39,13 +39,13 @@ Use `$CLAST_BIN` in place of bare `clast-plumbing` for all commands in this skil
 
 ## Step 1: Resolve the project
 
-If the user passed a slug as an argument (`/wakeup xesapps`), use it directly. Otherwise resolve from current working directory:
+If the user passed a slug as an argument (`/brief xesapps`), use it directly. Otherwise resolve from current working directory:
 
 ```bash
 $CLAST_BIN registry resolve "$(pwd)"
 ```
 
-If `pwd` doesn't resolve and no slug was given: print "Not in a registered project. Run `clast-plumbing registry add .` first, or invoke as `/wakeup <slug>`." and stop.
+If `pwd` doesn't resolve and no slug was given: print "Not in a registered project. Run `clast-plumbing registry add .` first, or invoke as `/brief <slug>`." and stop.
 
 ## Step 2: Gather data
 
@@ -73,7 +73,7 @@ $CLAST_BIN entries read <entry-path>
 Using the **synthesis prompt** (see below), produce a briefing of 2–5k tokens. Structure:
 
 ```
-## Wakeup briefing — <project>
+## Brief — <project>
 
 **Active thread:** <one-line from most recent entry's "Open threads" section, or "None">
 
@@ -101,11 +101,11 @@ End with one of:
 
 ## Step 4: Don't write anything
 
-Wakeup is read-only. Never invoke write-form subcommands (`entries write`, `breadcrumb '<text>'`, `snapshot`) from this skill.
+Brief is read-only. Never invoke write-form subcommands (`entries write`, `breadcrumb '<text>'`, `snapshot`) from this skill.
 
 ## Edge cases
 
-- **No entries for project**: print "No curated entries for `<slug>` yet. Run `/day-wakeup` to process recent sessions, or run `clast-plumbing sessions --project <slug>` to see what's available." and stop.
+- **No entries for project**: print "No curated entries for `<slug>` yet. Run `/wake` to process recent sessions, or run `clast-plumbing sessions --project <slug>` to see what's available." and stop.
 - **Slug resolves but no entries and no sessions**: print "Project `<slug>` registered but has no journal activity yet."
 - **Today's session count > 5**: summarize ("worked 12 sessions today, most recent 16:22 on branch `loop-guard-ngram`") rather than listing all.
 

package/{.claude-plugin/skills/day-wakeup → skills/wake}/SKILL.md

@@ -1,17 +1,17 @@
 ---
-name: day-wakeup
-description: 'Generate curated journal entries from yesterday''s Claude Code sessions across all projects. Use when the user says "/day-wakeup", "day wakeup", "morning briefing", "catch me up on yesterday", "what did I work on yesterday", "review my day", "process yesterday''s sessions", or otherwise signals they want to curate prior work across projects at the start of a new day. Runs `clast-plumbing snapshot` to ensure fresh data, then walks through each uncurated session from yesterday and proposes a draft entry the user can accept, edit, or skip. Prompts for promotion of decisions, common-issues, and workflows per accepted session. This is the once-per-day curation flow; for per-project briefings use /wakeup; for mid-session pivots use session-brief.'
+name: wake
+description: 'Generate curated journal entries from yesterday''s Claude Code sessions across all projects. Use when the user says "/wake", "wake", "morning briefing", "catch me up on yesterday", "what did I work on yesterday", "review my day", "process yesterday''s sessions", or otherwise signals they want to curate prior work across projects at the start of a new day. Runs `clast-plumbing snapshot` to ensure fresh data, then walks through each uncurated session from yesterday and proposes a draft entry the user can accept, edit, or skip. Prompts for promotion of decisions, common-issues, and workflows per accepted session. This is the once-per-day curation flow; for per-project briefings use /brief; for mid-session pivots use session-brief.'
 ---
 
-# Day Wakeup
+# Wake
 
 Process yesterday's Claude Code sessions across all projects. For each session, generate a draft journal entry and walk the user through accepting/editing/skipping it.
 
 ## Why this exists
 
-Curation at end-of-session has high friction (the user wants to stop, not summarize). Curation at start-of-next-day has lower friction (fresh eyes, easier to decide what's worth keeping). `/day-wakeup` is that start-of-next-day flow.
+Curation at end-of-session has high friction (the user wants to stop, not summarize). Curation at start-of-next-day has lower friction (fresh eyes, easier to decide what's worth keeping). `/wake` is that start-of-next-day flow.
 
-The transcripts themselves are captured automatically by the SessionStart hook + cron — the user never has to remember to log anything. What `/day-wakeup` does is **curate the captured transcripts into durable entries the user controls**, and prompt for promotion of decisions, common-issues, and workflows along the way.
+The transcripts themselves are captured automatically by the SessionStart hook + cron — the user never has to remember to log anything. What `/wake` does is **curate the captured transcripts into durable entries the user controls**, and prompt for promotion of decisions, common-issues, and workflows along the way.
 
 ## Step 0: Resolve the clast-plumbing binary
 
@@ -58,6 +58,27 @@ $CLAST_BIN --json sessions --since -30d
 
 Filter to sessions with `curated: false` or `stale: true` (stale sessions were curated but their transcript was updated since). If none remain, print "Nothing to curate — all sessions are curated or dismissed." and stop.
 
+### Step 2a: Auto-dismiss no-op sessions (deterministic, no LLM)
+
+Each session row carries a deterministic `substantive` flag (computed by
+`clast-plumbing` from the transcript, cached at snapshot time). A session is
+`substantive: false` when **Claude never replied** — an empty session, one that
+contains only slash commands like `/clear`, `/model`, `/config`, or one abandoned
+before any response. These are worthless to curate. (A session driven by a *custom*
+slash command still has assistant replies, so it stays `substantive: true`.)
+
+Before generating any drafts, auto-dismiss every uncurated session with
+`substantive == false` (skip this if the user set `CLAST_WAKE_AUTODISMISS_NOOP=0`):
+
+```bash
+$CLAST_BIN sessions dismiss <session-id> --reason "auto: no substantive content (empty / slash-command-only)"
+```
+
+Do **not** call the LLM for these. Dismissal is reversible via `clast undismiss <id>`.
+Remove them from the working set, keep a count, and report it in the final summary
+(e.g. "Auto-dismissed 5 no-op session(s)."). If nothing substantive remains after this,
+print "Nothing to curate — all remaining sessions were empty or slash-command-only." and stop.
+
 ### Triage when multiple days have uncurated sessions
 
 If uncurated sessions span more than one day (e.g., after a weekend or break), present a triage step before processing. Show a per-day breakdown:
@@ -113,15 +134,16 @@ For each session in the list:
    - **Accept** (any combination of accept-flavored options): pipe the draft to `clast-plumbing entries write` via stdin.
    - **Edit**: prompt the user for what to change, regenerate the draft incorporating their feedback, loop.
    - **Skip**: do not write.
-   - **Stop here**: end the entire `/day-wakeup` flow, leaving remaining sessions uncurated (user can resume tomorrow).
+   - **Stop here**: end the entire `/wake` flow, leaving remaining sessions uncurated (user can resume tomorrow).
 
 ## Step 4: Final summary
 
 After all sessions are processed (or user stopped early), print a summary:
 
 ```
-Day wakeup complete.
+Wake complete.
 Curated: 3 sessions across 2 projects.
+Auto-dismissed (no-op): 5 sessions.
 Skipped: 1 session.
 Remaining uncurated: 0.
 
@@ -130,15 +152,15 @@ Promoted:
   Common-issues: 0
   Workflows: 1
 
-Run `/wakeup <project>` to start working on a specific project today.
+Run `/brief <project>` to start working on a specific project today.
 ```
 
 ## Draft generation prompt
 
-The prompt templates live in `lib/clast/prompts/` so they are shared with the porcelain `clast wake` subcommand:
+The prompt templates are installed alongside the plugin under `$CLAUDE_PLUGIN_ROOT/lib/clast/prompts/`:
 
-- **System prompt:** `lib/clast/prompts/day-wakeup-draft-system.md`
-- **User prompt template:** `lib/clast/prompts/day-wakeup-draft-user.md` (uses `{{placeholder}}` syntax)
+- **System prompt:** `$CLAUDE_PLUGIN_ROOT/lib/clast/prompts/wake-draft-system.md`
+- **User prompt template:** `$CLAUDE_PLUGIN_ROOT/lib/clast/prompts/wake-draft-user.md` (uses `{{placeholder}}` syntax)
 
 When generating each draft, read those files and substitute the placeholders with session data. The user prompt template uses these placeholders: `{{project}}`, `{{branch}}`, `{{start}}`, `{{end}}`, `{{msg_count}}`, `{{first_turns}}`, `{{last_turns}}`, `{{breadcrumbs}}`.
 
@@ -156,7 +178,7 @@ After showing the draft, present:
   - `Accept + promote workflow` — also write a workflow file
   - `Edit` — user wants to revise; will prompt for changes
   - `Skip` — do not write this entry
-  - `Stop here` — end /day-wakeup entirely, leave remaining sessions uncurated
+  - `Stop here` — end /wake entirely, leave remaining sessions uncurated
 
 If `Skip` and `Stop here` are both selected, treat as `Stop here`. If `Edit` is selected alongside any accept option, treat as `Edit` first (the user wants to revise before accepting).
 
@@ -199,6 +221,7 @@ For promoted items (decisions, common-issues, workflows): currently these are tr
 ## Edge cases
 
 - **No uncurated sessions**: print "Nothing to curate — all sessions are curated or dismissed." and stop.
+- **All uncurated sessions are no-ops**: after Step 2a auto-dismisses them, nothing substantive remains — print "Nothing to curate — all remaining sessions were empty or slash-command-only." and stop.
 - **Multi-day backlog**: present the triage step (Step 2) so the user can choose scope before processing.
 - **`clast-plumbing snapshot` fails**: warn the user, then attempt to proceed with whatever's already in the manifest.
 - **`clast-plumbing show` fails for a specific session**: skip that session, note it in the final summary, continue with the rest.