The platform workflow's first run produced seventeen failures on Windows and macOS, where the JavaScript suite had never executed. Two were real defects in shipped code. A CONTAINMENT CHECK THAT REFUSED EVERYTHING ON WINDOWS `resolveWithinProfile` compared `target.startsWith(root + "/")` with a hardcoded forward slash. On Windows `resolve()` returns `C:\profile\file.txt` and the comparison asked for `C:\profile/`, so it matched nothing: document export and import refused every path, including legitimate ones. It failed CLOSED, so nothing leaked — the feature simply did not work. Now uses `relative()`, which needs no separator of our choosing. A COMMAND PARSER THAT ATE WINDOWS PATHS `parseCommandSpec` treated backslash as a POSIX escape unconditionally, so `--exec "C:\Program Files\tool.exe"` was spawned as the command `C:Program` with an argument `Filestool.exe`. Every Windows path handed to --exec or --stdin-to was mangled before the spawn. This is the SECOND instance of this exact bug today — the Python TUI palette had it too — which is what a whole class looks like rather than a one-off. `windows` is an explicit parameter, not a `process.platform` read, so both branches are tested from either platform. That parameter is the actual fix for the class: the reason these survived is that the Windows branch could only ever run on Windows. THE REST WERE TESTS, NOT CODE - `/bin/true` does not exist on macOS (it is /usr/bin/true), so a spawn failed with ENOENT and the refusal looked like the daemon's crash-safety fix having regressed. Now uses process.execPath, which exists everywhere by definition. - The oversized-request test treated its own EPIPE as fatal. The daemon caps the line, answers and hangs up; on macOS that hangup lands while the client is still pushing the remaining megabyte. Correct server behaviour was failing the test on one platform and passing on the other. - Three sink tests wrote `#!/bin/sh` files. Shebangs and /bin/sh do not exist on Windows. Now Node scripts run by the interpreter already running the tests. - Eight `0600` assertions and two clipboard tests are guarded by documented constants in test/helpers/platform.ts. Windows has no POSIX mode bits and the runner has no clipboard. The skips state a gap rather than closing it, and TODO.md now records that the Windows permissions question grew from two Python tests to ten across both languages — including the vault index and the semantic index, which hold secrets. With TypeScript becoming the only implementation that is a shipping question, not a test-suite one. Co-Authored-By: Claude Opus 5 (1M context) <[email protected]> Claude-Session: https://claude.ai/code/session_01Vtx94nXDmNgXEeThUZP4QN
seedpass-js
The SeedPass command-line interface, TypeScript port.
It reads and writes the same ~/.seedpass profiles as the Python
implementation — same derivations, same vault format, same Nostr sync — and
adds a session agent plus a scoped-token surface designed for AI agents and
automation.
Install
Requires Node.js 22 or newer.
# from a checkout
cd js && pnpm install && pnpm --filter @seedpass/cli build
node packages/cli/dist/seedpass-js.mjs --version
The build produces one self-contained file (dist/seedpass-js.mjs) plus a
.sha256 next to it. Copy it anywhere on your PATH; it has no runtime
dependencies beyond Node itself.
Quick start
export SEEDPASS_APP_DIR=~/.seedpass # omit to use the default
# Create a profile (mnemonic and password come from the environment,
# never from the command line)
export SEEDPASS_MNEMONIC="your twelve or twenty-four words"
export SEEDPASS_PASSWORD="your master password"
seedpass-js fingerprint add --name main
unset SEEDPASS_MNEMONIC # not needed again
# Unlock once per session through the agent
seedpass-js agent start &
seedpass-js vault unlock --ttl 900
seedpass-js entry add password github.com --length 20 --username you
seedpass-js entry reveal github.com
seedpass-js entry totp-codes
seedpass-js use github.com --clipboard
seedpass-js vault lock
Design: reference-first output
Default output never contains secret values. Entries are addressed by
reference (sp://entry/<id>), and secret-bearing fields are reported as
has_* flags:
$ seedpass-js entry get api-token
{ "id": "5", "ref": "sp://entry/5", "label": "api-token",
"kind": "key_value", "has_value": true, ... }
Plaintext leaves the vault only through commands that say so
(entry reveal, entry totp-codes, entry export-document,
util generate-password), or through a sink, which delivers a secret to its
destination without routing it through this process's output:
seedpass-js use api-token --exec ./deploy.sh # SEEDPASS_SECRET in the child env
seedpass-js use api-token --stdin-to "wc -c" # quote specs that contain flags
seedpass-js use api-token --clipboard
Scoped tokens for agents
An agent can be given a token that grants exactly what it needs and nothing else. Secrets are materialized inside the session agent, so the token holder never receives the seed and can never escalate to owner operations.
# As the owner:
seedpass-js agent token-issue --scope read use \
--kind key_value --label-regex '^api-' --uses 2 --ttl 600
# In the agent's process — no mnemonic, no password:
export SEEDPASS_TOKEN=<token>
seedpass-js entry list # metadata only
seedpass-js use api-token --exec ./deploy.sh # permitted
seedpass-js entry reveal api-token # denied: scope 'reveal' not granted
# Back as the owner — every grant, delivery and denial is recorded:
seedpass-js agent audit-verify
seedpass-js agent audit-tail -n 20
Environment
| Variable | Purpose |
|---|---|
SEEDPASS_APP_DIR |
Profile directory (default ~/.seedpass) |
SEEDPASS_MNEMONIC |
Parent seed; needed for profile creation, otherwise use the agent |
SEEDPASS_PASSWORD |
Master password, for fingerprint add and vault unlock |
SEEDPASS_TOKEN |
Scoped token; puts the CLI in token mode |
SEEDPASS_AGENT_SOCK |
Override the agent socket path |
SEEDPASS_SECRET |
Set by use --exec in the child process only |
Compatibility
Interoperability with the Python implementation is verified on every change
by scripts/cross_impl_check.py: each side opens the other's profiles,
backups round-trip, edits are mutually visible, conflict merges agree, and a
snapshot published by Python restores here with secrets intact. See
docs/typescript_port_compatibility_matrix.md.
Not yet ported: PGP RSA keys (not byte-reproducible; ed25519 is at parity),
and the semantic and api command groups.