Metis

Load an agent's system prompt, contract, skill and project contexts from the RC.

Log a completed agent run to the database for audit and dashboard tracking.

Retrieve recent agent run history from the database.

Scrub PII from text and return anonymized version + replacement map.

Output rail: check a drafted response for leaked PII before it's sent.

Read a sensitive data file and return a REDACTED preview (masked values).

Return a unified diff comparing original and anonymized text.

Append a row to the consent_ledger table.

Retrieve recent consent events from the audit ledger.

Set the current network access policy for all agents.

Return the current network access policy.

List the knowledge background packs available to install, and what is already installed.

Install a knowledge background pack: fetch its documents and register the layer.

Write a pack manifest for one of your installed background layers.

Create a timestamped backup of the Metis SQLite database.

AES-256-GCM encrypt a backup file.

Decrypt a .enc backup produced by encrypt_backup().

List all backup files with size, age, and checksum availability.

Run SQLite integrity_check on a backup file.

Restore the Metis database from a backup.

Configure the nightly backup schedule.

Return the current backup schedule configuration.

Run one turn of a brainstorm session, returning relevant context.

Retrieve all turns in a brainstorm session.

List recent brainstorm sessions with title, turn count, and status.

Freeze a brainstorm session as a saved Markdown output.

Save a script, snippet, reusable function or template to the Code Repository.

Record a dataset's data dictionary — one stored entry per variable.

Record one cleaning/transformation step for a dataset (its lineage).

Search the Code Repository for prior code, variables and treatments.

Assemble the raw material to write a NEW script from previous work.

Return the full Metis user configuration from user-config.yaml.

Add a specialist context to the user profile, or update it if it exists.

Activate or deactivate a specialist context.

List all user contexts (general + specialist) with active status.

Write the full user-config.yaml produced by the first-run config wizard.

Write user-preferences.json produced by the first-run config wizard.

Import an ideas document (Word/text/markdown) and capture each idea.

Delete the .first-run marker file to signal that the config wizard is complete.

Fill a Today-surface board (Outbreaks, Events or Funding) with items you found on the web.

Fetch RSS feeds and add new items to news_briefs.

Scan inputs/literature/ for new PDFs and register them in literature_metadata.

Scan the inbox/ folder and auto-transcribe any audio files to ideas.

Retrieve recent news briefs from the database.

Run all Metis update scans in sequence and return a combined report.

Start a new course build pipeline.

Save an approved course outline after Step 2 (Scope Plan).

Return the current status of a course build (or all active builds).

Step 3 — Save harvested source metadata for a course.

Step 4 — Save the approved curriculum design for a course.

Step 5 of the course build — save one drafted lesson to disk.

Step 6 — Run quality checks on a drafted course before publishing.

Finalise and publish a completed course build.

Register a data automation trigger — "when X happens, do Y".

List all data automation triggers with their status.

Update a data trigger (enable/disable, rename, change schedule).

Delete a data trigger permanently.

Manually execute a data trigger right now.

Profile a tabular dataset: shape, dtypes, null %, unique counts, distributions.

Profile a dataset and return specific recommended cleaning operations.

Apply cleaning operations to a dataset and write a new file.

Compare two dataset profiles and produce a side-by-side diff.

List supported dataset formats and installed library versions.

Pull decisions out of session summaries into the open-decisions ledger.

Walk the decisions waiting on you, most-repeated first.

Close an open decision: agreed, rejected, deferred or dropped.

Make an authenticated API call to the configured DHIS2 instance.

Query DHIS2 metadata with a simplified interface.

Return ONE earned, not-yet-shown feature tip for the current moment (or '').

Return the tiny first-run orientation — the 3-5 highest-value capabilities.

Adjust the feature-tips preference (the user's control).

Report the discovery-tips state and a simple adoption read.

Run a one-screen health check on Metis.

Scan all tracked files and report which have changed since last scan.

Add a single file to the tracked-files list and start watching it.

Register all relevant files in a project folder so Metis can read them.

Read the content of a file and return it as text.

List files in a folder.

List files in the Metis basket (legacy & inspiration documents).

Move a basket item to a stable project folder (promote from basket to active storage).

Stop tracking a file so Metis no longer watches it for changes.

Extract and index full text from all PDFs in the Metis library.

Full-text keyword search across all indexed PDFs.

M5.7.1 — Probe external tool result for injection patterns.

M5.7.3 — Load the Metis constitutional policy for agent context.

M5.7.4 — Validate a sub-agent output before passing to the next pipeline stage.

Generate a portable session handoff brief.

Store an idea in the SQLite ideas table.

List captured ideas from your knowledge base, newest first.

Store a journal entry with auto-extracted mood and energy.

Append to (or read) today's daily note — one rolling note per day.

List journal entries from your knowledge base, newest first.

Retrieve all contacts from the contacts table.

Add or update a contact record.

Retrieve all glossary terms.

Add a glossary term, or update its definition if it already exists.

Search library, meetings, and news for items related to given text.

Find cross-domain connections for given text.

Assemble context from multiple sources for brainstorming.

Generate an image using AI and save it to the PKM.