Skip to content

Security: teamaltevo/agent-hours

Security

SECURITY.md

Security and public deployment

Public repository versus private data

The intended public tree contains application code, synthetic tests, lockfiles, documentation and example deployment settings. It must not contain:

  • local/ settings, classifier caches, private session indexes or imported exports.
  • Generated reports/, conversations, source databases, commands or tool output.
  • .env files, authentication stores, private keys or real deployment settings.
  • worker/wrangler.jsonc, .dev.vars or .wrangler/ state.

The package files allowlist separately excludes those paths from distribution. The OAuth client ID in the source is a public client identifier, not a secret. The configured Worker URL and submission phrase are deliberately public. Git author names and email addresses are normal repository metadata and may also become publicly visible; review them before changing repository visibility.

The public phrase does not authenticate reports

altevo-agent-hours is visible in source. Anyone can send a schema-valid report with it. It does not prove identity, organization membership, TimeSentry sign-in or accurate measurements. Treat emailed reports as unverified submissions.

The Worker cannot be used to choose arbitrary recipients or forward arbitrary attachments. It validates allowlisted aggregate fields and builds attachments itself. Input sizes, an approximate short-window limiter, deduplication and a D1-backed daily email budget limit abuse. They do not eliminate denial of service, Cloudflare resource consumption, false reports or exhaustion of the email budget. Add identity-based authentication or human verification if those risks are unacceptable.

Consent and outbound data

The CLI discloses the report receiver before the run proceeds. Classification and sign-in default to Yes at the requested [Y/n] prompt; users can cancel with n. The default report receiver is enabled without key setup. Use --no-send to generate local reports without submission.

Only completed, validated aggregate reports are uploaded. Source conversations, command text, tool output, private session IDs, paths and credentials are excluded. Aggregated work hours, dates and activity measurements can still be sensitive. Email recipients and the mail infrastructure necessarily receive that aggregate data.

Delivery diagnostics are deliberately separate from report content. Worker logs contain only stage/error enums, random request UUIDs, opaque receipt hashes, elapsed time, byte sizes and HTTP status. Names, email addresses, headers, payloads, commands and paths are not logged; automatic invocation logs and traces are disabled. CLI transport diagnostics follow the same restrictions.

Reports also include the person's display name when TimeSentry provides one, or a name derived from their own returned @altevo.ca address when the profile name is absent. This is an explicitly allowed personal-data field; raw email addresses, employee IDs and the rest of the profile remain excluded. The name and its source are self-reported to the public Worker, not proof of identity.

Project-level reports explicitly include catalogue project/customer display names, daily project aggregates and API-equivalent token-cost estimates. The consent screen discloses these fields before collection/submission. Raw project IDs, aliases, Jira links, unknown model strings and filesystem-derived names are not exported. Public pricing model IDs come only from the vendored tariff allowlist. Labels are validated and HTML-escaped; project CSV labels are quoted and spreadsheet-formula prefixes are neutralized. These estimates are unverified chargeback-planning data, not signed billing records. Project names and costs must not be written into transport diagnostics or Worker logs.

TimeSentry OAuth tokens remain in memory and are not forwarded to the report Worker. The earlier report-upload token is obsolete; it must not be published or confused with the deliberately public phrase.

Scope of review

Automated tests and targeted tracked-file/history scans are useful checks, not a formal security audit or a guarantee that no vulnerability or sensitive datum exists. A public repository also needs a review of its sharing defaults, metadata, third-party dependencies and organizational approval to disclose its source.

There aren't any published security advisories