causly-server

Quality and hardening plan

This folder is the engineering plan for making Causly Server enterprise quality. It is internal working documentation for contributors. User-facing docs live on the docs site (server.causly.in/docs).

File	What it holds
P0.md	Must be done before the 1.5.4 release and the 10 Nov 2026 launch
P1.md	Next release (1.6): deeper hardening, real approval flow, broader test coverage
P2.md	Later: larger features, long-running verification, enterprise extras

Every item has an ID (P0-03), the evidence (file and line), the impact, the fix, and an acceptance test. An item is done only when its acceptance test exists and passes.

Where this came from

• An independent audit run on 30 Oct 2026 against master at commit c6bc132.
• A re-check of every audit claim against the code and npm audit on 31 Oct 2026.
• The open GitHub issue #2 and the “Limitations” page of the docs.
• A QA review of what the existing test suite (56 test files) does not cover.

Audit claims that did not survive the re-check are listed at the bottom of this page, so they are not worked on by mistake.

Rules of the sprint

1. Test first. For every security item, write the failing test that reproduces the problem, then fix it. The test stays as a regression test.
2. One item, one commit. Keep each commit small so it can be reviewed and reverted alone.
3. Work on a branch. P0 work goes on security/hardening-1.5.4. master is untouched until the branch is green on Ubuntu, macOS and Windows.
4. No blind fixes. Reproduce first. If an item cannot be reproduced, write down why and close it.
5. Docs move with code. Any change to a tool, a permission level or a default updates the matching page in the docs site in the same change set. The pages live in the agenv repo under website/docs/server/. Each P0 and P1 item has a “Docs to update” line that names them.
6. No new unrequested scope. New ideas go into P1 or P2, not into the sprint.

Definition of done for a release

• npm audit --omit=dev reports no high or critical issues
• npm test passes on Ubuntu, macOS and Windows (Node 22 and 24)
• Every registered tool has an explicit permission level (enforced by a test)
• npm pack --dry-run shows only the allowlisted files
• CHANGELOG.md has a security section for the release
• causly-server doctor --test passes on a clean machine with real keys
• Package published with provenance, tag pushed, MCP Registry entry updated

Timeline to launch (10 Nov 2026)

Dates	Focus
1 to 3 Nov	P0-01 to P0-06: dependencies, issue #2, redaction, shell injection, SSRF, file guards
4 to 5 Nov	P0-07 to P0-09: CI, permission levels, MCP conformance and annotations
6 to 7 Nov	P0-10 to P0-12: fuzz tests, install and upgrade matrix, release hygiene
8 Nov	Release 1.5.4, ask Glama and the other directories to re-scan
9 Nov	Buffer. Renew the Google token (causly-server login google), final doctor --test
10 Nov	Product Hunt launch

README and ROADMAP are updated last, after P0 is green.

Limitations page to plan

Limitation (docs)	Plan
Issue #2: npm install fails inside nested cmd.exe on Windows	P0-02
Workspace limit is a guardrail, not a sandbox	P1-03 (Docker mounts, resources), documented honestly in the meantime
Approval gate is a flag, not a screen	P1-01 (real operator approval)
Commands inherit the server environment, including keys	P1-04
Recurring calendar events: updating changes one instance only	P1-12
Google token expires after 7 days in Testing mode	P1-13 (production publish is a user-side step)
Transcription needs the internet	P2-01 (local engine)
No sync between machines	P2-05
BoundaryAttest is a proof of concept	P2-04
Direct cloud tools (Azure, GCP, AWS)	Out of scope
LinkedIn, X and WhatsApp outreach	Out of scope

Audit claims that did not hold up

• Master key validation is weak. It is already checked: the key must decode to exactly 32 bytes (tools/secretsOps.js). No work needed beyond a unit test.
• Test tools bypass the approval gate. run_tests, run_lint, run_typecheck and run_build run the project’s own scripts. That is what they are for, so this is a policy decision, not a bug. See P1-05.
• CVE numbers in the audit. The audit quoted CVE IDs that npm audit does not report. npm audit reports GitHub advisory IDs (GHSA). Work is tracked against npm audit, not the quoted IDs.
• “410+ tests”. That figure came from an older pull request. The suite has 56 test files and CI is green, but the count has to be taken from a fresh run.

This site is open source. Improve this page.