v0.5.3 — Windows support, contributor fixes, bounded objective commands

Windows support, the latest community fixes, and objective commands that are bounded in memory and time. Added: the plugin runs on Windows, and CI now runs the full suite on Linux, macOS and Windows for Python 3.10, 3.11, 3.12 and 3.14 (content removal without directory descriptors, file identities SQLite can hold, real time zones through tzlocal/tzdata, a number-based initiative picker, LF checkouts); a project can keep up to 4,000 characters of a failed command objective's stderr (verification.objective_detail_chars; the default is unchanged). Fixed: Install from Git keeps the Desktop app (#23, #25); "Match Hermes" follows Hermes Desktop's theme through its documented --ui-* variables (#22, #24); inside a named profile every profile is listed (#18); the test suite never reaches the machine's live Kanban boards or profiles. Security: objective commands run under one deadline (60 s by default) and keep only the start and end of each output stream (64 KiB each), so a flood keeps memory flat; a count on the last line is now measured even after a long output, so the result of a very chatty count objective can change. At the deadline or on interrupt the command is stopped with what it started: on Linux and macOS its process group, then any helper that detached (found through a private marker in the command's environment); on Windows its process tree, with taskkill /T run from System32. What can still escape is listed in the threat model. An independent security review checked the command runner before release. Thanks to Andrew Peltekci (@appletechie) for the reports and pull requests behind these changes (#18, #19, #22–#25). Gate: pytest 1,902 passed / 3 skipped, browser 336/336, dashboard 73/73; native lane against the pinned Hermes revision green. No schema change.

v0.5.2 — the catalog-review fixes, restored and verified on the ladder codebase

Correction release. v0.5.0/v0.5.1 (the ladder and the contributions wave) did not carry two of the external review's must-fixes — they lived on a review branch that was never merged into the ladder work. This release restores them and re-verifies everything: Fixed: the Kanban host no longer writes process-global HERMES_KANBAN_HOME (context-local override only; board root from Hermes core); Desktop "Match Hermes" reads the theme through the plugin SDK's useTheme(); guarded Kanban transfers assign through core's assign_task; "Evaluate now" honours the command-evaluator autonomy gate; workflow revisions check added steps' assignees and bind the proposer to the authenticated principal; the popup poller is torn down on pagehide and PDF previews render sandboxed; the __V4_DISABLE/dockyard:test:deeplink test hooks are removed. Gate: pytest 1,841 passed / 2 skipped, browser + dashboard suites green. No schema change.

v0.5.1 — community contributions wave

First community-contributed release: integration of @appletechie's PRs #15/#16/#17/#20/#21 (full) and #19 (patch 1), each rebased onto v0.5.0 through the full gate with authorship and CHANGELOG credit kept. See CHANGELOG [Unreleased] for the feature-by-feature detail.

v0.4.0

The Project Stewardship Desktop app becomes the main surface, and every part of the plugin now shares one name: hermes-project-stewardship.

https://github.com/Sahil-SS9/hermes-project-stewardship/releases/download/v0.4.0/walkthrough.mp4

Highlights

• A Desktop app for all your projects: Dashboard, Projects, Roadmap, Backlog, Workflows, Approval Inbox, Notifications and App settings, with a project switcher that remembers your last choice.
• Workflows your bots follow: 30 built-in templates, a builder with an AI sidebar that suggests first, loops with limits, if-stuck steps, and runs that move on by themselves.
• You approve, with the evidence in front of you: every decision waiting on you sits in the Approval Inbox. Notifications pop up, and optional Discord alerts go through the Discord bot your Hermes already runs.
• Project memory and chat: each project keeps its own rules, decisions and facts for the bots working on it, and you can open a Hermes chat with its lead or any of its bots.
• Archive and restore exactly: an archived project disappears from every list, count and inbox, and Restore, from the project switcher, brings it back as it was.

Upgrading from 0.3.0

1. Export a backup first: stewardctl --db <your database> export --output <folder>.
2. Switch the plugin on once under its new name (#4): hermes plugins enable hermes-project-stewardship, or in Hermes Desktop. Your data stays where it was.
3. The first time 0.4.0 opens your database it moves the schema from 17 to 41 and keeps every row. Never open an upgraded database with 0.3.0; to go back, restore your export into a new database with 0.3.0.

Full upgrade notes and every change: CHANGELOG.md.

Staying on 0.3

The last 0.3 build stays available on the release/0.3 branch. To install or go back to it, pin its commit:

hermes plugins install Sahil-SS9/hermes-project-stewardship --force --ref 4f733a81fa5f9da964e253e577b036b0b9608b85

If 0.4.0 has already opened your database, restore your pre-upgrade export into a new database first (step 3 above): 0.3 must never open an upgraded database.

Fixed

• #1: the tests collect from a checkout folder of any name.
• #4: one name everywhere, so switching the plugin on mounts its dashboard backend.
• #5: no exact version pins, so Hermes can install the plugin next to its own pinned tools. Thank you @apoapostolov for the fix (#6).
• Projects Hermes does not know ("unlinked") no longer break a page, and the API answers them with a clear 404.
• Two parts of Hermes opening an older database at the same moment both come up.

Checked before release

hermes verify on the release commit: 1,663 Python tests, 321 browser tests and 73 dashboard tests pass, plus 1,658 tests against an unmodified Hermes checkout (0.21.6). The upgrade was rehearsed on a copy of a real 0.4.0-development database: every row kept, and the old version refuses the new schema instead of writing to it.