Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion .github/workflows/ci.yml
Original file line number Diff line number Diff line change
Expand Up @@ -58,7 +58,7 @@ jobs:
strategy:
fail-fast: false
matrix:
suite: [api, browser, production]
suite: [api, browser, production, calibration, orchestration]
steps:
- uses: actions/checkout@d23441a48e516b6c34aea4fa41551a30e30af803 # v6
with:
Expand Down
58 changes: 53 additions & 5 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,11 @@ SyberLabs' job-search product: a private review workspace that remembers the rol
- Exact draft acceptance: you approve specific wording; changing it requires review again.
- Interview notes and follow-ups that do not reset a submitted or live-loop application.
- Import a tracker CSV through column mapping and a record preview. Imported statuses stay research; existing Relay status and accepted drafts are preserved.
- A verified fact ledger and learned style rules that gate agent drafts at the API. Unsupported claims are refused, not queued.
- Batch review of logged drafts. Staging a draft does not accept it or send an application.
- Choose between real postings to set preferences, then plan the week to maximise the expected value of the best single offer.
- Pull public Greenhouse and Lever boards onto existing job identity. Discovery arrives Held.
- Record outcomes with a receipt before a submission counts toward reply rates. Ended applications cannot be reopened by import.
- Explore fictional example records; no real applicant data is included.

## Integrations
Expand All @@ -24,8 +29,10 @@ SyberLabs' job-search product: a private review workspace that remembers the rol
- **[Claude](integrations/README.md)**: Draft preparation from your verified facts through the Claude API, using your own key.
- **[Grok Bot](integrations/GROK_BOT.md)**: Validated research and draft file exchange inside the Bot's VM.
- **[Tracker CSV](integrations/README.md)**: Map columns, preview rows, and import a local tracker file as research.
- **[Greenhouse](integrations/README.md)**: Read-only public board pull onto existing job identity. No credentials and no application sending.
- **[Lever](integrations/README.md)**: Read-only public board pull onto existing job identity. No credentials and no application sending.

You stay in control. Integrations are explicit file and command handoffs. Every returned draft needs human review. Relay does not send applications or sync accounts in the background.
You stay in control. Integrations are explicit file and command handoffs. Every returned draft needs human review. Agents cannot accept a draft or submit an application. Public board pulls are read-only. Relay does not send applications or sync accounts in the background.

ChatGPT uses a prompt and file handoff; Codex also supports a local CLI adapter. This release does not include a hosted ChatGPT app or MCP connection.

Expand All @@ -41,7 +48,7 @@ Relay is the workspace that remembers. Import research as evidence. Review the a
- **Research keeps its history.** Matching posting URLs join the existing opportunity; changed source notes remain separate observations. Rediscovery preserves an existing interview or submitted status.
- **Acceptance belongs to the text.** An imported Ready label does not approve a new draft. Accept it in Relay; changing the accepted text requires another review.
- **Handoffs belong to a job and version.** Draft packets and editor checks reject stale work instead of silently replacing a newer review.
- **Bring the tools you already use.** ChatGPT, Codex, Claude, Grok Bot, Notion, Obsidian, and a tracker CSV can supply research and drafts. Relay keeps the review record across those handoffs.
- **Bring the tools you already use.** ChatGPT, Codex, Claude, Grok Bot, Notion, Obsidian, a tracker CSV, and public Greenhouse or Lever boards can supply research and drafts. Relay keeps the review record across those handoffs.

Relay records acceptance inside its workspace. It does not send applications, verify every claim, or prove which words were submitted to an employer. URL matching cannot identify every repost across different job boards.

Expand Down Expand Up @@ -84,6 +91,22 @@ Choose **Prepare for ChatGPT** or **Prepare for Codex** in **Connect your tools*

The [GrokCell folder](grokcell/README.md) includes First Principles, Product Ideation, Red Flag, and Garbage Collector. These templates do not connect to Relay or grant access to application records. The [Relay adapter](integrations/GROK_BOT.md) remains the guide for exchanging job research and drafts. See also [context management](context-management/README.md). This is a pinned copy of the MIT-licensed [GrokCell project](https://github.com/sdcarlson/grokcell).

## Autonomy and review

Relay separates what an agent may do freely from what only you may do. The split is enforced in the API, not in a prompt.

The **fact ledger** (`/profile`) holds claims you have verified. Paste resume text to propose candidates; extraction writes nothing and marks nothing verified. A fact can carry an expiry for anything that goes stale, such as a current title or a headcount.

The **style card** holds the voice rules learned from your corrections. The agent reads both through `relay_read_profile` and never writes either.

A writing agent logs drafts through `relay_log_draft` at whatever volume it likes. Each sentence that asserts something checkable must cite a verified fact id, and every number must appear in a cited fact. A draft that fails either rule is refused at the API and is never written, so an invented achievement cannot reach a review queue. An agent short of a fact is expected to log with `confidence: "low"` rather than guess.

Logged drafts accumulate until review is worth your time (`/review`). A session opens on the first draft in an unseen role cluster, on style drift within a proven one, on enough low-confidence drafts, or on a full batch — in that order. The session groups repeated habits so one decision covers several drafts, flags near-identical letters, and turns your corrections into style rules. Closing a session with no rules changes nothing about the next batch.

Autonomy is then earned per role cluster. After enough reviewed drafts come back close to unchanged, a cluster graduates and its later drafts are placed into the matching job record unattended. Staging is not acceptance: status is untouched, `accepted_draft` is not set, and accepting an exact draft remains a human action in the workspace. A correction or an expired fact returns the cluster to full review.

Relay still does not send applications. Nothing here submits, emails or messages anyone.

## Run locally

Node 24 and pnpm are required. From the repository root:
Expand All @@ -101,6 +124,8 @@ pnpm build
pnpm exec wrangler d1 execute DB --local --config dist/server/wrangler.json --persist-to .wrangler/state --file drizzle/0000_sticky_robbie_robertson.sql
pnpm exec wrangler d1 execute DB --local --config dist/server/wrangler.json --persist-to .wrangler/state --file drizzle/0001_careless_leader.sql
pnpm exec wrangler d1 execute DB --local --config dist/server/wrangler.json --persist-to .wrangler/state --file drizzle/0002_job_key_observations.sql
pnpm exec wrangler d1 execute DB --local --config dist/server/wrangler.json --persist-to .wrangler/state --file drizzle/0003_profile_calibration.sql
pnpm exec wrangler d1 execute DB --local --config dist/server/wrangler.json --persist-to .wrangler/state --file drizzle/0004_selection_and_outcomes.sql
pnpm dev
```

Expand All @@ -127,9 +152,27 @@ curl -sS http://127.0.0.1:8787/api/workspace \

The Sites plugin on `pnpm dev` strips caller-supplied `oai-authenticated-user-*` headers and injects the mock user only after the sign-in cookie. Header mock is for the built Worker. Production must use the Access gateway from the delivery foundation.

## Selection and outcomes

Drafting well is not the largest lever. Which jobs you apply to is, because reply odds vary by orders of magnitude with fit while letter quality varies perhaps twofold. These stages address that.

**Preferences** (`/preferences`) are elicited by forced choice over real postings, because people cannot state trade-off weights in the abstract but choose between two concrete jobs instantly. Around a dozen comparisons fit the five weights well enough to rank a large pool. Until you answer, Relay has no opinion and says so rather than inventing one.

**The plan** (`/plan`) maximises the expected value of the *best single offer*, not the sum over applications, because you accept one job. That is an expected maximum, and computing it exactly produces a spread of odds on its own: once a likely offer is held, further similar roles add almost nothing and a long shot starts winning the comparison. No reach/match/floor ratio is hardcoded. Selection is greedy on gain per minute under the attention budget you set, and every row carries the reason it earned its place.

**The read plane** pulls public Greenhouse and Lever boards through `board-pull`, normalising onto the same `jobKey` identity the workspace already uses. It reads only, needs no credentials and no account, and every posting arrives Held — discovery never implies a decision.

**Outcomes** (`/track`) close the loop. Terminal states are local records of what actually happened and are deliberately not valid import statuses, so the existing import behaviour is unchanged and rediscovery can never reopen an application that has ended. A submission recorded here requires a receipt: the confirmation URL, reference or email subject. Imported submissions are kept and shown, but they arrive without a receipt and are excluded from the reply rates, because counting a send that may never have happened would corrupt every estimate built on it.

Reply rates are reported as intervals with the evidence count that produced them, and they are descriptive. Samples are small, the market moves and a job search cannot be run as an experiment, so nothing here establishes that a change caused an outcome.

**Interview preparation** falls out of the citation graph rather than being a separate feature: because every claim had to cite a verified fact, Relay already knows what each application commits you to defending.

Relay still does not submit applications. There are no write-plane adapters in this release: nothing here fills in a form, sends an email, or messages anyone, and recording a submission is you telling Relay what you already did.

## Integration boundaries

Connectors are local commands and explicit file import/export. Notion and Claude use your credentials. ChatGPT is a prompt/file handoff; Codex also supports a local CLI adapter; Grok Bot uses a documented command/file adapter. There is no automatic background sync, autonomous hunting, application sending, or LinkedIn messaging. Generated claims still require review. Production deployment and independent user adoption remain separate from this local workflow.
Connectors are local commands and explicit file import/export. Notion and Claude use your credentials. ChatGPT is a prompt/file handoff; Codex also supports a local CLI adapter; Grok Bot uses a documented command/file adapter. Agents may log and, for a graduated cluster, stage drafts; they cannot accept one or change an application status. Board pulls read public endpoints only and use no account identity. A recorded submission is your own report of something you did elsewhere and needs a receipt. Citation enforcement checks that a claim traces to a fact you verified; generated text still requires review. There is no automatic background sync, autonomous hunting, application sending, or LinkedIn messaging. Production deployment and independent user adoption remain separate from this local workflow.

See [Local assisted browser trial](REVIEW.md#local-assisted-browser-trial) for the earlier `3b1efd7` read-only Notion, hand-transcribed Bot file, and browser acceptance, reload, and reimport evidence, still shown on a read-only reopen at `59ec7ac`.

Expand All @@ -144,13 +187,18 @@ pnpm test
pnpm exec tsc --noEmit
pnpm lint
pnpm public-copy:check
node tests/api.test.mjs
pnpm test:calibration
pnpm test:orchestration
```

The project, organization, personal profile, and website share one [maintained product description](docs/public-copy.json). [Public-copy process](docs/PUBLIC-COPY.md) explains generation, checks, and automatic refreshes.

API checks need a running local server and write only fictional test records. Connector tests mock vendor responses; they do not prove live account access. Domain, import, editor, and API tests cover status preservation, duplicate matching, exact acceptance, stale edits, and authentication rejection.
`pnpm test` is the unit gate. The API, calibration, and orchestration live suites each get a dedicated freshly migrated database via `pnpm test:api`, `pnpm test:calibration`, and `pnpm test:orchestration`. Do not run calibration and orchestration against the same D1: calibration graduates clusters and retires facts. Connector tests mock vendor responses; they do not prove live account access.

For an existing local database still on migration 0001, apply 0002, then 0003 and 0004. From 0002, apply 0003 then 0004. From 0003, apply only 0004. Migration 0003 adds the fact ledger, style rules and draft-review tables. Migration 0004 adds preference, choice and outcome tables and defaulted posting columns on `jobs`; existing rows stay. Imported Ready is research evidence; a new record stays Held until its exact draft is accepted in Relay.

For an existing local database still on migration 0001, apply 0002 from the setup commands. Imported Ready is research evidence; a new record stays Held until its exact draft is accepted in Relay.
Domain, import, editor, profile, selection, and API tests cover status preservation, duplicate matching, exact acceptance, citation refusal, terminal outcomes that cannot be reopened by import, stale edits, and authentication rejection.

## Hosting and privacy

Expand Down
Loading
Loading