kuhy 2026-11-11, multi-day research without re-setup.
W1: once a window was used in a lock, the block's purge spares
~/.local/share/Steam (Librarian's install and local saves) for the
rest of the lock; ~/steam, ~/.steam and the rest still go. The
package stays removed, hosts/iptables are re-applied, and since the
kept install could otherwise be launched without the package, every
Steam game is killed on sight. A record that cannot be read also
keeps the folder (an irreversible delete never rests on unknown
state); the block itself stays fully applied.
W2: the 48 h cap is gone; --until is bounded by the lock's end. An
open window may be extended ONCE, to the lock end at most, never
shortened, never reopened after it ended. The extension is a
separate extended_until field beside until, so a daemon still on
the previous code reads the original window instead of rejecting the
record as corrupt (which would have purged the folder).
Failure modes unchanged: missing/corrupt record = no window, and a
corrupt one refuses a new window or extension.
Checked: 3325 tests at 100% coverage; pylint 10/10 in parts; vmbox:
extension accepted once, refused beyond the lock end, earlier or
equal, second time and non-root; extended end honoured past the
original end; at close the Steam dir and a save survive, ~/steam and
~/.steam go, a game started after close is killed; extension to
exactly the lock end accepted. Not verified on the host.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWWzuCEg1m87jpKyRrw
kuhy 2026-11-11: game-design research needs Librarian (app 4197610)
actually played during the 6-day total block. New root command
total-block-research --until T opens ONE window per lock: <= 48 h,
never past the lock end, a rerun never extends it.
The command only validates and records (/var/lib/steam-backlog-
enforcer, root 0700; lock id + deadline + two latches). The daemon
opens and closes everything: Steam hosts names, the store block, the
total-block chain (IP cache kept) and the steam package, reinstalled
with only guard-lib's package-block hook masked for that one pacman
call (--hookdir, proved in a guest). Lock file, hook and end date are
never touched. Browser-game sites stay null-routed.
Only Librarian may run: loop-local allowed set, strict mode (no
whitelist exceptions, no zero-achievement games), launchers and
counted games killed; state.json is never read or written.
Expiry latches the record, then the ordinary tick re-applies hosts
(now reconciled each tick), chain, uninstall and purge; the normal
store block is restored once, unconditionally.
Failure modes: missing/corrupt/unparseable record = no window and a
corrupt one also refuses a new window (fail closed); a clock set
back cannot reopen a latched window.
Checked: 3290 tests at 100% coverage; whole-tree pylint 10/10 run in
three parts (the hook exceeds the 60 s cap, so SKIP=pylint here); vmbox
guest with real guardctl package-block and the real unit, window
opened, Librarian-only enforced, clock moved past the deadline and the
full block re-applied. Not verified on the host.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWWzuCEg1m87jpKyRrw
Requiring 100% locked the user onto one game for weeks, which proved
ineffective. The assignment and every manual pick now release as soon as one
achievement is unlocked after the game was assigned (Steam unlock_time vs the
pick's started_at / State.current_assigned_at).
- finished_app_ids still means 100% complete; a game released below 100% gets
released_at plus a 7-day cooldown and returns to the pool.
- Next pick: least recently assigned first, then shortest HLTB time.
- Releasing never evicts: released/expired picks stay installed and playable
until the user explicitly chooses another game (new pick-manual, or
accepting the next assignment); with no replacement the current game stays.
- Tampering detection judges a released game by unlocks after its release.
- The daemon reloads every State field each tick so its sweep save cannot
overwrite cooldowns or assignment times written by the CLI.
- State.load backfills assignment times once and persists them; load-time
migrations moved to _state_migrations.py.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PhzaTAWtRfJk8gre6NLSka
Expose the backlog enforcer's read surface and two gated write actions as
typed MCP tools (steam_backlog_enforcer._mcp) so Claude Code and its
subagents can query and drive the backlog without shelling out.
- _mcp.py: FastMCP stdio server. Read tools get_dataset/get_status/
get_stats/list_backlog; gated writes pick_manual/block_gaming (preview
unless confirm=True). Logs to stderr; never exposes Config/the API key.
- _actions.py: stdout-free, state-only cores shared by CLI and MCP
(apply_manual_pick, status_payload, is_manual_pick_locked). cmd_pick_manual
now delegates its state mutation here.
- .mcp.json + scripts/setup_mcp.sh: project-scoped registration via a
dedicated ~/.venvs/steam-backlog-mcp interpreter; mcp is an optional-dep,
kept out of the systemd/CLI system-python path.
- pyproject: fix setuptools flat-layout discovery and declare the previously
undeclared direct deps aiohttp + websockets, un-breaking the editable install.
- 100% coverage of the new modules; server verified end-to-end over stdio.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LQGQj8Ne6TriBExv91nG8v