Find a tool.
Make something happen.

Anything to do with finishing an audio file the user already has -- .wav, .flac, .aiff, .mp3. Reach for it when the ask sounds like "make this sound finished", "this is too quiet for YouTube", "get rid of the harsh S sounds", "the room sounds boxy", "make this podcast match last week's episode", "level these three files to each other", "tighten the low end", "hit -14 LUFS without clipping", "trim the silences", "get rid of the umms and ehms", or "tighten up this recording". It measures what is actually there (loudness, true peak, spectral balance, crest factor, sibilance, ambience), FINDS things in it (onsets, silent spans, filler words with their timings), then runs an EDITING and MASTERING chain over it: cut listed regions, strip or shorten silences, de-ess, de-verb, parametric EQ, EQ-match against a reference recording, MULTIBAND compression and multiband dynamic range control, saturation, controlled ambience, loudness targeting and true-peak brickwall limiting. Every CHAIN STAGE (cut, strip-silence, gate, expand, deess, dereverb, eq, eq-match, compress, saturate, reverb, stretch, pitch, loudness, limit) appends to a plan, and one render applies the whole chain in a single pass, so a chain can be inspected and re-run, and the whole job is ONE shell command rather than a round trip per stage. The read-only/reporting verbs (analyze, detect, verify, check, config, manifest) and the plan-lifecycle verbs (plan, preset) do not append to a plan themselves -- see the verb table below for which is which. This is mastering and cleanup, NOT mixing: it works on a finished stereo or mono programme, not on multitrack stems. It uses speech recognition internally to locate filler words, but it does NOT produce transcripts -- do NOT use it for video files, for transcription as an output, or for music generation.

Manifest name: aud

• Measure a file honestly before touching it -- LUFS, true peak, crest factor, spectral balance
• Bring a podcast or music file to a loudness target without clipping or inter-sample peaks
• Trim the silences out of a spoken-word recording, or shorten them to a fixed length
• Trim the pauses without chopping the start of a word -- edits are padded, snapped to a safe cut point and crossfaded, rather than falling wherever the detector put the boundary
• Get rid of the "umms" and "ehms" and long hesitations in an interview or a voice-over
• Tighten up a rambling recording -- find the pauses and fillers, cut them, master the result
• Find where the onsets are in a recording, to decide where an edit should land
• Tame harsh sibilance on a vocal or spoken-word recording
• Reduce a boxy or reverberant room on a recording made in the wrong space
• Make one recording sit in the same tonal balance as a reference recording
• Extract the EQ curve of a recording you like and apply it to another file
• Even out dynamics per frequency band with multiband compression, not one blunt full-band squeeze
• Retime or re-pitch a programme without changing the other
• Verify that a finished render actually meets the loudness and ceiling it was asked for

Snapshot refreshed: 2026-09-22T03:32:37Z

Source commit: f36a4d9e5cc8787b50ae107282dabc549c58dd2c

Ref: main / Path: .

Researches a question across many sources and returns a short brief plus the citations behind it. Reach for it when the ask sounds like "what do we actually know about X?", "find me sources on this", or "I need to decide this and have not read anything yet". Searches the live web and synthesises, returning a brief plus a pointer to the full evidence kept on disk. Do NOT use it to check specific claims you already have -- that is fact-check -- or for questions answerable from the code or documents already in front of you.

Manifest name: deep-research

• Find out what is known before committing to a decision
• Get a short sourced answer without reading the sources first
• Build a durable evidence record for later questions
• Produce a bibliography without collecting references by hand

Snapshot refreshed: 2026-09-22T03:32:38Z

Source commit: 7540598affad363882571587f3bc20b968cd7305

Ref: main / Path: tools/deep-research

Stands up isolated, realistic environments from a declarative profile so software can be tested as though actually deployed, and manages them for their whole life: author a profile, launch it, run commands inside, copy files in and out, update it, and tear it down. Use when "tests pass locally" is not enough evidence and you need to exercise code the way a real deployment would.

Manifest name: amplifier-digital-twin-universe

• Turn a description of what you want to test into a launchable environment profile
• Test a web app in a container that mirrors its real deployment
• Verify a CLI tool installs and runs cleanly from scratch
• Simulate an end-user environment without touching production
• Exercise code against mocked third-party services without changing its configuration
• Work out why a host cannot launch environments, and what to run to fix it

Snapshot refreshed: 2026-09-22T03:32:39Z

Source commit: 3913c70532f5a640b1968f66c99029ce7d4fbfd8

Ref: main / Path: .

Takes things someone has asserted and checks each one against evidence, returning a verdict per claim -- supported, refuted, unverifiable or opinion -- with the sources each rests on. Reach for it when the ask sounds like "is any of this actually true?", "check the claims in this draft before it goes out", or "where did that number come from?". Claims are checked INDEPENDENTLY, so one false claim does not condemn the rest of a document. Do NOT use it for an open question with no claim in it yet -- that is deep-research -- or to check code against its tests.

Manifest name: fact-check

• Check the claims in a draft before it goes out
• Find which assertions hold and which merely sound right
• Audit a single verdict down to its sources
• Re-check claims against evidence a previous research run gathered

Snapshot refreshed: 2026-09-22T03:32:39Z

Source commit: 7540598affad363882571587f3bc20b968cd7305

Ref: main / Path: tools/fact-check

Suggests a prepared read or list action using a bounded local model call. Use when a harness has eligible workspace targets and wants a fast advisory choice with explicit abstention.

Manifest name: amplifier-fast-decisions

• Choose among caller-validated workspace read targets
• Measure a local decision scorer independently of an agent harness
• Record advisory decisions alongside parent and child session metadata

Snapshot refreshed: 2026-09-22T03:32:40Z

Source commit: ff00b285a28bb51c6180d06888afad26b713bd39

Ref: main / Path: .

Finds the most-starred GitHub repositories created in a recent window, optionally filtered by language, as structured data -- and can rank them against a stated goal using a model. Also manages the authenticated user's account: list your own repositories, create a new one, change a repository's visibility, list pending repository invitations, and accept or decline them. Use when you want to know what is gaining traction, which of several projects fits what you are trying to do, or you need to inspect or change your own GitHub repositories and act on pending collaboration invitations.

Manifest name: github-repos

• Find the fastest-growing repositories in a language right now
• Shortlist candidate libraries for a specific job
• Check whether a problem space already has a popular solution
• List your own repositories, filtered by visibility or affiliation
• Create a new repository (private by default) for yourself or an org
• Change a repository from private to public, or back
• Check for pending repository invitations
• Accept or decline a pending collaboration invitation

Snapshot refreshed: 2026-09-22T03:32:40Z

Source commit: aaac263e5c29807de89490362353cc0a5561fd23

Ref: main / Path: github-repos

Reads and sends Gmail, searches and creates Google Contacts, and reads and creates Google Calendar events, all as structured data. Use when the task is about email, inbox triage, contacts lookup, or calendar/schedule questions for a Google account.

Manifest name: gmail

• Summarize or triage an inbox (list unread, list by sender, list by label)
• Read one email's full body and reply to it in-thread
• Send a new email on the user's behalf
• Look up a contact's email/phone/organization, or create a new contact
• Answer "what's on my calendar" / "any meetings today" and create new events

Snapshot refreshed: 2026-09-22T03:32:40Z

Source commit: aaac263e5c29807de89490362353cc0a5561fd23

Ref: main / Path: gmail

Fetches the current Hacker News front page as structured data, and can produce a themed digest of those stories using a model. Use when you want what the technical community is reading right now, either raw or summarised.

Manifest name: hacker-news

• Get the current Hacker News top stories as structured data
• Produce a themed digest of what the front page is about today
• Check whether a topic is currently being discussed on Hacker News

Snapshot refreshed: 2026-09-22T03:32:40Z

Source commit: aaac263e5c29807de89490362353cc0a5561fd23

Ref: main / Path: hacker-news

Safely analyze Home Assistant data and directly operate registered household services under revocable owner-granted trust.

Manifest name: ha-analysis

• Summarize the structure of caller-supplied Home Assistant evidence without network access
• Inspect explicitly named Home Assistant entities through a bounded read-only interface
• Check a configured Home Assistant API or consent to a bounded entity-registry display search
• Configure one Home Assistant origin and store its long-lived access token in the Linux operating-system secret store
• Enable revocable connection-bound trust, discover registered services, resolve explicit targets, and invoke household actions
• Run an embedded model-backed household operator through the same trusted control boundary

Snapshot refreshed: 2026-09-22T03:32:41Z

Source commit: f741cb13292f39639141f1573484567beec866d4

Ref: main / Path: .

Works with Spotify through a bounded model-backed `do` workflow, a reviewable `plan` then deterministic `apply` workflow, and deterministic catalogue, playlist, library, device, and playback commands. `do` exposes the closed, supported music-domain library surface (not general tools), with `--read-only`, `--no-playback`, and one-observation `--local` effect boundaries. Each invocation is currently fresh and ephemeral; named create/resume sessions are not yet implemented, and no latest session is selected. Use deterministic commands for exact ID-targeted work.

Manifest name: music-deck

• Carry out one explicit playlist brief, with bounded search, read-back, and playlist writes
• Turn a brief into a readable plan that a person can review before deterministic application
• Search Spotify, inspect playlists or saved items, and explicitly edit playlists through deterministic commands
• List authenticated Spotify API devices and separately observe local Spotify Connect advertisements on Linux
• Control playback only on an eligible Spotify API device through deterministic commands

Snapshot refreshed: 2026-09-22T03:32:43Z

Source commit: ca35796b44fb465db27d122d219d519c43b3b49d

Ref: main / Path: .

Extract local video ranges as MP4, GIF, or PNG with inspectable plans and receipts. Find remembered moments through Amplifier Agent with bounded local evidence. Optional local dashboard for reviewing, refining, exporting and reopening moments.

Manifest name: outtake

• Make a shareable clip or reaction GIF from a known local time range
• Extract a still with literal custom text
• Revise a cut while preserving earlier plans and exports
• Find a remembered moment using captions, clues and permitted source frames

Snapshot refreshed: 2026-09-22T03:32:45Z

Source commit: 98a9454c79d2aa3c9e5c4c9bdb0436f8f39520f2

Ref: main / Path: .

Explore visual alternatives for an app, choose an experience, and refine a standalone prototype before production development.

Manifest name: possibly

• Compare meaningful task flows before committing to an app design
• Refine a selected experience while preserving deliberate choices
• Hand a self-contained interactive prototype and decision record to a development agent

Snapshot refreshed: 2026-09-22T03:32:47Z

Source commit: 12d41ea8861d6bcc4efb3aa202553f00d47a1d9e

Ref: main / Path: .

Performs bounded web demonstrations, navigation-only by default, and returns continuous application footage with step evidence. Use for prepared demo dashboards, with optional exact Stories comment authority; not generation, login or post-production.

Manifest name: showrun

• Record a prepared pitch deck without taking over human browser tabs
• Demonstrate an isolated installed Stories revision through its actual dashboard
• Inspect a prior take without repeating model calls or target interactions

Snapshot refreshed: 2026-09-22T03:32:48Z

Source commit: 1e9e950fe08686a1d6ab13e396cd53fa5759a919

Ref: main / Path: .

Creates, validates, and evaluates smart tools that follow the Amplifier Smart Tool Spec. Use when you want to package domain expertise as a smart tool, check that an existing one conforms to the spec, or measure how well its smart capabilities work.

Manifest name: smart-tool-creator

• Scaffold a new smart tool that is ready to push and conforms to the spec
• Check an existing smart tool against the spec and its conformance kit
• Build and run evaluations for a smart tool's model-backed capabilities

Snapshot refreshed: 2026-09-22T03:32:49Z

Source commit: 88854e007a5a36872f6b86499a78656fe5f5ff6a

Ref: main / Path: .

Controls Spotify playback (play, pause, skip, volume), searches the catalog, browses playlists and devices, and launches the desktop app. Use when the task is about music, songs, playlists, Spotify, or playing/pausing/searching audio.

Manifest name: spotify

• Answer "what's playing" / "what song is this" and report progress
• Play a specific track by search query, or resume paused playback
• Pause, skip, or adjust volume on the active playback device
• Search Spotify's catalog for tracks by artist, title, or album
• Browse the user's playlists or list available playback devices
• Launch the Spotify desktop app when no device is available

Snapshot refreshed: 2026-09-22T03:32:49Z

Source commit: aaac263e5c29807de89490362353cc0a5561fd23

Ref: main / Path: spotify

Create evidence-based presentations and documents with agent highlights and anchored comments. Use for source-backed communication and shared review.

Manifest name: amplifier-smart-tool-stories

• Turn supplied evidence into an HTML presentation or structured document
• Review a story with a person and highlight material that needs attention
• Answer or act on anchored feedback while preserving the reader's place

Snapshot refreshed: 2026-09-22T03:32:51Z

Source commit: 82c8b8f866d28a26a52705807bfc83d27209b833

Ref: main / Path: .

Reads the Team Pulse lens API -- a knowledge base mined for a team (decision/meeting wikis, code/repo wikis, roster, reflection questions) -- as structured data, plus a model-backed answer that retrieves and cites its sources. Use for factual questions about a team's decisions, what shipped, who owns what, why something happened, or how something works.

Manifest name: team-pulse

• Answer "what did we decide about X" / "who owns Y" / "why did we do Z" with citations
• Search or browse a team's mined corpus (decision wikis, code/repo wikis)
• Look up the team roster or a reflection question by id
• Bulk-download the corpus for offline analysis (embeddings, grep, your own agents)
• Record a session-mined answer to a reflection question

Snapshot refreshed: 2026-09-22T03:32:51Z

Source commit: dc98f879b8a831edc7c21c9cb64bd27d67c224db

Ref: main / Path: .

Tells you what is happening across every tmux session on a machine - which are parked at a prompt, which finished and how, which need a human - and, only under explicit per-invocation confirmation, types into one or starts one. Reach for it when a box is running many tmux sessions (agents, builds, REPLs, long jobs) and "what needs me right now?" is the question, or when something needs one keystroke or one new session delivered safely. The model-backed verbs run through amplifier-agent's engine embedded in-process.

Manifest name: tmux-fleet

• Survey every tmux session on a machine and see which are parked at a prompt
• Triage a busy fleet to find which sessions plausibly need a human, and why
• Read one session's scrollback with an honest bound on what was captured
• Interpret what a session's recent output means without attaching to it
• Type a command into a session, or start a new one, under explicit confirmation

Snapshot refreshed: 2026-09-22T03:32:52Z

Source commit: 5655d7ef507a6453f72b94ee016fa38df666cb61

Ref: main / Path: .

Create, review, reuse and deliver motion graphics using embedded Amplifier Agent. Retain editable source, encoded output, frame evidence and revision history.

Manifest name: unfold

• Explain a technical or mathematical idea through animated geometry and text
• Revise an explanation while retaining the previous version
• Review, rename and export saved outputs without a model

Snapshot refreshed: 2026-09-22T03:32:54Z

Source commit: e9fa1ee1dedd4d33a26d41ade4a29df931f5fe84

Ref: main / Path: .

Anything to do with a video file the user has - .mp4, .mov, .mkv, .webm. Reach for it when the ask sounds like "cut this down to the bit where she explains pricing", "add captions to this", "make this shorter", "stick these three clips together", "speed up the boring middle", "put some music under it", "where does he mention the deadline?", or "make this match our brand colours". Trims and cuts, joins clips with transitions, retimes and ramps speed, zooms, burns in captions, removes/replaces/mixes audio, grades colour or matches a reference image, vignettes, writes and speaks a narration fitted to the video's own timing, finds a moment by what was SAID or SHOWN, and verifies a finished render. Chain the verbs with pipes - the whole edit is one ffmpeg pass. Do NOT use for images, audio-only files, or downloading video.

Manifest name: vid

• Trim a video down to one section, or cut a section out of the middle
• Join several clips together, with or without a transition between them
• Speed a recording up, slow it down, or ramp the speed across a stretch
• Burn captions into the picture from a subtitle file
• Find the moment someone said something, or the moment something appeared on screen
• Replace, remove or mix the audio, or lay a music bed under a demo
• Match a video's colour to a reference image, or apply a named look
• Write a narration from a prompt and speak it so it fits the video's own timing
• Check that a finished render is actually what was asked for

Snapshot refreshed: 2026-09-22T03:32:55Z

Source commit: 6640efce4de422d517a6025683fba34db5d926fa

Ref: main / Path: .

Produces safe Microsoft 365 briefings and retrieves workplace context through the official Work IQ MCP server. Use it for daily preparation, meeting research, and permission-aware access to the signed-in user's work data.

Manifest name: amplifier-smart-tool-workiq

• Prepare a prioritized daily briefing from Microsoft 365 work context
• Gather relevant context before a meeting
• Create reusable domain-specific workplace intelligence workflows
• Ask Microsoft 365 Copilot a workplace question through a stable CLI
• Fetch a bounded, explicitly selected Microsoft 365 resource
• Discover Microsoft 365 resource paths and inspect their fetch schemas
• Validate and execute reviewable read-only Microsoft 365 data plans

Snapshot refreshed: 2026-09-22T03:32:57Z

Source commit: b3fc85d0509f30ba66e45f552a91f313e26de0b4

Ref: main / Path: .