Clipboard Memory

Use this skill to recover clipboard history from the local clipmem archive. The OpenClaw-native packaging variant lives at extras/openclaw/clipboard-memory/; the Hermes Agent-native package lives at extras/hermes/clipboard-memory/; the portable packaging mirror lives at extras/agent-skills/clipboard-memory/.

Clipboard Memory

Use this skill when the target is likely something the user copied on this Mac, even if they never say "clipboard". Prefer it over web, repo, or filesystem search only when the item likely lived on the clipboard.

Use this skill when

The user asks things like:

• "what was that command I copied?"
• "show me the URL I copied from Safari earlier"
• "find that snippet, path, note, or link I copied yesterday"
• "give me the exact text I copied, not a summary"
• "what did I copy before I restarted?"
• "paste me back that SQL I was looking at"
• "get the PDF I copied last week"
• "show me everything I copied from Xcode today"

Also trigger when the user asks indirectly to recover or paste back something from earlier and the clipboard is the most plausible source, even if they never say "clipboard", "copy", or "paste".

Do not use this skill for

• web search or current-events lookups
• searching the repository, local files, or browser history when the user never copied the target
• content the user typed but never copied to the clipboard
• anything on a non-macOS machine (clipmem only captures NSPasteboard)

Fast decision workflow

1. Confirm the target likely came from the clipboard.
2. If empty results would be surprising, run scripts/check-setup.sh first. Prefer scripts/check-setup.sh --json when you want structured diagnostics. The prose fallback is references/setup-check.md.
3. Pick the narrowest starting command:
   • vague recollection -> recall
   • chronological history / "today" / "yesterday" / "in order" -> timeline
   • recent unique things -> recent
   • exact phrase, punctuation-heavy string, URL fragment, or file suffix -> search --mode literal
   • snapshot id already known -> get
   • binary / image / PDF recovery -> get then export
4. Add only the filters the user actually implied: --hours, --app, --kind, --has-url, --has-file-url, --prefer-recent.
5. If the first pass is weak, empty, or low-confidence, broaden once before giving up: widen --hours, drop source filters, inspect alternatives, then switch between recall, search, recent, and timeline.
6. Return the recovered content plus provenance such as observed_at, app_name, and snapshot_id when useful.

Capability map

The repo-side agent-native action parity contract lives in docs/action-parity.md. Use it when you need the maintained map from user-visible outcomes to agent-accessible commands, entity CRUD expectations, and derived-cache boundaries.

Critical behaviour rules

• Before answering from a stale, empty, or ambiguous archive, run clipmem agents context --format json and use generated_at, health, settings, app state, recent activity, revision, stats, privacy, and capability fields to decide whether to broaden search or diagnose setup.
• Always use --format json when you will parse the response. --format toon is for token-efficient enumeration only. --format jsonl is for streaming many rows into a pipeline. Never parse md or text.
• Treat recall as a convenience ranking helper, not an authority. For uncertain cases, compose primitive commands in this order: search, recent, timeline, get, then OS follow-through such as pbcopy, open, or open -R.
• Never claim "nothing found" until you have broadened the search once and checked truncated / next_cursor.
• When best_match_confidence is "low" or there are several plausible hits, present the top candidates instead of pretending certainty.
• For exact-text requests, quote best_text verbatim. Do not paraphrase commands, SQL, code, URLs, or file paths unless the user asked for a summary.
• For binary-only snapshots where best_text is empty, say so plainly and recover the raw bytes with export; do not invent text.
• Keep the query and filters stable while paginating with --cursor.
• Prefer recent over timeline when the user wants unique things rather than every copy event.

Health check and setup

Before querying, confirm the setup is healthy — otherwise empty results may be a stale watcher, not a true miss:

1. Background capture must be running. clipmem setup is the canonical fix; Homebrew users can also use brew services start clipmem.
2. The binary clipmem must be on PATH with access to ~/Library/Application Support/clipmem/clipmem.sqlite3.
3. Run scripts/check-setup.sh when results look wrong. It exits 0 on a healthy host, 1 if the watcher is stale, 2 if the binary is missing, 3 if clipmem doctor or clipmem service status fails. scripts/check-setup.sh --json emits structured status for agents. The prose equivalent is in references/setup-check.md.

Command ladder

Always pick the narrowest command that answers the question.

1. clipmem recall — best-first ranked answer with alternatives. Start here for most "what was that thing I copied" requests.
2. clipmem timeline — chronological capture events, including repeated copies of the same content. Use for "today", "yesterday", "before I restarted", "in order", or "every time".
3. clipmem recent — recent unique snapshots, deduplicated by snapshot.
4. clipmem search — direct lexical / FTS matching. Use --mode literal for exact substrings or punctuation-heavy content.
5. clipmem get SNAPSHOT_ID — nested item / representation detail for a single snapshot already in hand.
6. clipmem export SNAPSHOT_ID --item N --uti UTI --out PATH — raw bytes for binary, image, or PDF payloads.
7. clipmem ocr candidates, clipmem ocr get, clipmem ocr clear, and clipmem storage image-candidates — inspect queued OCR or image optimization work before running batch workflows, or clear one stale OCR result.
8. clipmem settings reset --format json — reset capture policy and ignored apps when the user explicitly asks to restore defaults.
9. clipmem service providers --format json — inspect service provider state without starting or stopping capture.
10. clipmem service revision --format json — inspect archive revision counters without probing service providers.
11. clipmem app settings, clipmem app launch-at-login, clipmem app update-check run, or clipmem app quit with --format json — inspect or change menu bar app preferences and app-owned state when the user asks about app defaults, update checks, or quitting the app.
12. clipmem agents context --format json — compact health, settings, app state, recent activity, revision, stats, privacy, and capability context before multi-step work.

Primitive command taxonomy

Primitive commands expose one bounded read or mutation that can be composed directly. Convenience workflows such as recall, setup, purge, ocr run, and storage optimize-images remain useful, but verify uncertain results with search, recent, timeline, or get, and preview broad mutations with candidate or dry-run commands when available.

The full flag reference, JSON envelope, and kind values live in references/commands.md, references/json-schema.md, and references/examples.md.

Common recovery playbooks

• Command, SQL, or code snippet — start with clipmem recall "..." --format json --limit 5; if punctuation matters, follow with clipmem search "..." --mode literal --format json.
• URL, path, or filename fragment — add --has-url or --has-file-url; scope by --app safari or another app only if the user implied it.
• "Everything I copied today / from Xcode" — use clipmem timeline --hours 24 --app xcode --format json (or --hours 48 for "yesterday", --hours 168 for "last week").
• Recent unique clipboard items — use clipmem recent --hours N --format json or --format toon when you only need a compact list.
• Image or PDF recovery — locate the snapshot with recall or search, inspect it with get, then recover bytes with export.
• Before a restart or long ago — start broader than you think (--hours 72, 168, or more) and prefer timeline if ordering matters.

Output format rule

• --format json — structured output. Retrieval envelopes are stable within schema_version: 2; management and inspection commands use command-specific JSON shapes, so parse documented keys directly.
• --format toon — flat, token-efficient list. Prefer for high-cardinality enumeration (timeline, search, recent, recall) when you only need the top fields. Note: get does not support toon.
• --format jsonl — newline-delimited records. Use when streaming many rows into a pipeline.
• --format md / --format text — human-readable previews only; never parse these.

--json is an alias for --format json on search, recent, timeline, get, service revision, capture-once, and doctor.

Reading the response

Read these JSON fields first; walk nested items[].representations[] only after a get call:

• best_candidate.best_text — the flattened primary text.
• best_candidate.urls — URL array (empty when none).
• best_candidate.file_paths — file-URL array.
• best_match_confidence, why_selected, and alternatives — confidence and fallback options.
• observed_at, app_name, and kind — provenance and content shape.
• next_cursor, truncated — pagination state.
• schema_version — pin to 2 for stability.

Full schema in references/json-schema.md.

Quick examples

# best-first answer
clipmem recall "that command I copied" --format json --limit 5

# Safari today, chronological
clipmem timeline --app safari --hours 24 --format json --limit 25

# recent unique items, token-efficient
clipmem recent --hours 72 --format toon --limit 20

# exact URL or punctuation-heavy string
clipmem search "launchctl bootstrap" --mode literal --format json --limit 25

# health check with structured output (from the skill directory)
./scripts/check-setup.sh --json

# recover an image
clipmem get 42 --format json
clipmem export 42 --item 0 --uti public.png --out ./clipboard.png

Troubleshooting

If recall looks empty or weak, widen --hours, drop source filters, or switch to timeline, recent, or search. For setup issues, sandbox PATH problems, or binary-only snapshots, see references/troubleshooting.md.

Exit codes

0 success · 1 uncategorized runtime · 2 invalid args · 3 not found · 4 unsupported format · 5 database error · 6 platform error.