manage-account

Files (1)

• SKILL.md

SKILL.md (first lines)

# Manage Account

Use this skill when the user wants to **change something** in their Omnimatch account: edit a profile, refine search criteria, create private scoring entities from text / links / uploaded files, manage inbox filters, act on a match (like / reject / respond / view link / unlock / share contact), or upgrade/downgrade their plan. The Omnimatch MCP server exposes only a single **codemode** tool ([Cloudflare Codemode](https://developers.cloudflare.com/agents/api-reference/codemode/)) — call it with a `code` parameter (async arrow function body) that runs the appropriate `codemode.<tool>(args)` calls.

**Important:** Existing profile edits are text-first. To refine an existing entity's description or search criteria, use `codemode.update_entity_criteria({ entityId, text })` — the server stores the text as a scoring file internally and regenerates the profile. For creating a new **private scoring entity** from an uploaded file, use the private upload workflow below.

**Confirmation pattern:** Destructive or credit-spending tools (`reject_match`, `unlock_match`, `reevaluate_match`, `remove_entity_files`, `remove_inbox_filter`, `update_entity_title`, `update_entity_external_link`, `respond_to_match` with `direction: "less"`) require `confirmed: true`. The first call returns `{ requiresConfirmation: true, action, details, message }` — show the action to the user and only re-call with `confirmed: true` once they confirm.

**When talking to the user:** Always use **real titles**, never IDs. Refer to entities by their **title**, markets by **market title**, matches by the row's **title** (the other side's name), types by the **entity type title** (e.g. "job", "candidate"), and plans by **plan name**.

## Entity edits

- Title → `codemode.update_entity_title({ entityId, title, confirmed })`.
- External link → `codemode.update_entity_external_link({ entityId, externalLink, confirmed })` (pass `null` to clear).
- Email frequency → `codemode.update_entity_email_notifications({ entityId, frequency: "never" | "best_only" | "all" })`.
- Search criteria refinement → `codemode.update_entity_criteria({ entityId, text })`.
- Add a note → `codemode.add_entity_note({ entityId, name, text })`.
- Remove files → `codemode.remove_entity_files({ entityId, fileKeys, confirmed })` (keys from `get_entity` files[]).

## Private Scoring Entities

Use these when the user wants to upload a private opportunity/profile to score against their own existing entities. Private scoring entities are visible only to the owner and are excluded from cross-owner search.

- Text or link private upload → `codemode.create_private_entity({ marketId, entityTypeTitle, title, externalLink?, initialCriteriaText?, confirmed })`.
- List existing private scoring entities → `codemode.list_private_entities({ marketId, entityTypeTitle? })`.
- Rescore existing public profile against private scoring entities → `codemode.score_against_private_entities({ entityId, confirmed })`.
- Delete a private scoring entity → `codemode.delete_private_entity({ entityId, confirmed })`.

### Private File Upload Workflow

1. Call `codemode.get_private_entity_file_upload_url({ marketId, contentType, name })`.
2. Upload the file bytes with PUT to the returned `uploadUrl`.
3. Call `codemode.enqueue_private_entity_file_upload({ marketId, entityTypeTitle, title, stagedKey, contentType, name, confirmed: true })` using the returned `stagedKey`, `contentType`, and `name`.

The queued job creates the private scoring entity, attaches the file privately, and runs profile generation/scoring in the background.

## Inbox filters

Inbox filters are saved free-text questions answered per match. Each filter becomes its own inbox keyed `query:{filterId}`.

- Create → `codemode.add_inbox_filter({ entityId, prompt, strictness: "lenient" | "normal" | "strict" })`. Plan-gated (Match insights). Returns `{ filterId, inboxKey }`.
- Update → `codemode.update_inbox_filter({ entityId, filterId, prompt?, strictness? })`.
- Remove → `codemode.remove_inbox_filter({ entityId, filterId, confirmed })`.

## Match actions

Match actions all take `{ matchId, entityId }` (the match id and your entity id):

- Approve / "more like this" → `codemode.like_match({ matchId, entityId })`.
- Reject / "less like this" → `codemode.reject_match({ matchId, entityId, confirmed })` — confirm before calling.
- Free-text feedback → `codemode.respond_to_match({ matchId, entityId, direction: "more" | "less", textFeedback?, confirmed })` — `direction: "less"` requires confirmation.
- Get just the external link → `codemode.view_match_link({ matchId, entityId })`. Returns `{ externalLink, locked, message? }`.
- Unlock a credit-gated match → `codemode.unlock_match({ matchId, entityId, confirmed })` — deducts one credit; confirm before calling.
- Share contact details with an on-platform match → `codemode.share_my_contact_with_match({ matchId, entityId, name?, email?, externalLink?, confirmed })` — at least one of email/externalLink required. Only meaningful for owner-to-owner matches. **Do not call this for externally-sourced matches**: there's no owner on the other side to receive the share, and the public contact is already on `get_match.otherEntityContact`.
- Re-evaluate a match's score → `codemode.reevaluate_match({ matchId, entityId, confirmed })` — confirm before calling. Useful after the user updates their profile and wants the score to reflect the new context. Costs one re-evaluate credit on credit-enabled markets; fails if either side's profile isn't ready (missing scoring files, no extracted criteria, or marked outdated). The re-evaluation runs asynchronously and **can take up to 5 minutes** to complete; while it's pending, `get_match` may show `yourCriteriaScore`/`theirCriteriaScore` as `null` and inbox-filter answers as `computed: false`. Tell the user up front so they aren't surprised, and don't aggressively poll — wait at least a few minutes between checks.

## Match discovery (manual search)

Trigger a fresh discovery run with `codemode.discover_matches({ entityId, timeframe, searchSpace, totalConnections })`. This is the manual equivalent of the automatic search whose schedule you read via `get_next_match_discovery` (account-info).

**What it does — and what it doesn't:**

- It **finds and adds new matches**. It searches other entities in the market (internal) and the web (external), then creates fresh connections. It **excludes profiles you're already connected to or have rejected**, so it never duplicates or re-scores matches you already have.
- It is **not** a re-score. To refresh the score on an existing match after the user edits their profile, use `reevaluate_match` (single match). The platform also re-evaluates outdated matches automatically when criteria/files change — so "I updated my profile, will a search re-score my matches?" → no, a search brings in *new* matches; the existing ones re-score on their own (or via `reevaluate_match`).
- Before searching it **clears stale matches**: connections older than 60 days that were rejected or never acted on by either side are removed. Liked/approved and recent matches are kept.

**Cost & timing:** On credit-enabled markets a manual search **spends a search credit**. This tool has **no `confirmed` flag**, so tell the user it will run a search (and use a credit where the market charges for it) and get a yes before calling. The run happens in the **background** (a workflow); results land in the inboxes when ready — typically a few minutes. Don't poll aggressively.

**Parameters** (all optional, with defaults):

- `timeframe`: `"1 day" | "1 week" | "1 month" | "1 year" | "no limit"` — recency window for external web results. Default `"1 week"`.
- `searchSpace`: `"external only" | "internal only" | "both"`. Default `"both"`.
- `totalConnections`: `"auto"` or a number 1–50 — cap on how many matches to bring back. Default `"auto"`.

## Subscription / plan changes

- List plans → `codemode.list_plans({ marketId })` for plan name, description, isFree, isCurrentPlan, price (live from Stripe), features, limits.
- Change plan → `codemode.change_plan({ marketId, targetPlanId })` returns `{ billingPortalUrl, currentPlanId, targetPlanId, message }`. Open the URL — Stripe handles upgrade/downgrade/cancellation. No confirmed flag needed; Stripe is the confirmation surface.

## Workflow patterns

- **"Reject this match"** — call `reject_match` first without `confirmed`; show the preview to the user; if they confirm, call again with `confirmed: true`.
- **"Unlock that match"** — same pattern; warn that it costs one credit before re-calling with `confirmed: true`. Private scoring entity matches are already unlocked.
- **"Add a filter for senior roles"** — `codemode.add_inbox_filter({ entityId, prompt: "Is this a senior role?", strictness: "normal" })`. After it returns the filterId, tell the user the filter has been queued for output-format inference and will appear in the inbox once matches are evaluated.
- **"What does my filter say about this match"** — `codemode.list_match_inbox_filter_answers({ matchId, entityId })` (read-only — also available in account-info).
- **"Run a search" / "find me new matches now"** — `codemode.discover_matches({ entityId })`. If the market charges credits, tell the user it spends a search credit and confirm first (no `confirmed` flag exists). Explain it runs in the background and new matches will appear in the inboxes in a few minutes, and that it adds *new* matches rather than re-scoring existing ones.
- **"Upgrade to the Pro plan"** — `codemode.list_plans({ marketId })` to find the plan with name "Pro" and its planId; then `codemode.change_plan({ marketId, targetPlanId })`. Hand the user the `billingPortalUrl`.

From omnimatch/omnimatch-mcp, MIT.

More from omnimatch/omnimatch-mcp