Coordinator updates #152

Open opened by topos-backsword.test.pds.did.bot

VS Code's agent host drops the coordinator's self-started turns: replies after background tasks, and scheduled wake-ups (microsoft/vscode#332073, fix in review as #336753). So those updates never reached the maintainer.

Until that's fixed, the coordinator posts a short comment here whenever a wake-up has news: what merged, what's ready for review, and your next action. Quiet wake-ups post nothing.

🤖 Generated with Claude Code

Update, 2026-11-10 afternoon

Self-merged: #136 planning (heist planning before each mission; no-plan contracts unchanged), #145 rig connections gated by traits (Electronics and Improviser), #142 NPCs (guards and staff; bluff, bribe, tailgate, knock out), #132 rig and ammo, #133 doors.

New PRs, in blocker review, which I'll self-merge if clean:

  • #149: no floats. Hex maths is all integer and boolean, and an architecture rule now forbids f32/f64 in game code.
  • #150: physical impediments: shutters, lasers, safes, vents and rubble.
  • #151: lifepath starting kit. A runner holds their background's gear and keycards.

Waiting on you: merge #138, the definitions design note, which now records your answers (DAG-CBOR with CIDs, no floats, runtime definitions can leave a mission). #141, the registry crate, follows it.

Bug filed: #146, loop-smoke is flaky on main.

🤖 Generated with Claude Code

Update, 16:25

You merged: #138 (definitions note), #137 (devlog), #149 (no floats; hex is integer-only) and #151 (starting kit). Their plan boxes are ticked (#153, housekeeping).

In flight:

  • #141, the registry crate: blocker review running, and I'll self-merge it per #138.
  • #150, physical impediments: rebasing over your merges.
  • Digital impediments: being prepared on top of #150, and will open once it lands.
  • Loadout in planning: being built.

Nothing needed from you right now.

🤖 Generated with Claude Code

Update: I self-merged #141, the definition registry crate: atproto JSON packs, DAG-CBOR, CIDs, floats rejected. The review found no blockers, and it's ticked in plan/defs.md (#154). The first data slice has started: traits and door routes become an embedded pack, and the debug menu gets 'Load definitions' and 'Spawn kind here'.

🤖 Generated with Claude Code

Update: I self-merged #150, physical impediments: security shutters, laser tripwires, safes, bolted vents and rubble, each with several ways past. The review found no blockers. Floor generation re-checks that every floor can be finished, and it's at ruleset 11. One small bug is filed as #156 (a breaker overlapping a staff NPC on tutorial floor 1).

Next: digital impediments open as a PR now that #150 is in. #155, loadout in planning, is getting a 'bring nothing' fix, then review.

🤖 Generated with Claude Code

Update: I self-merged #155, loadout in planning. You pick which held gear to bring, the mission's rig is built from exactly that, and bringing nothing gives a zero-tech run. Unplanned missions are unchanged. The review found no blockers, and the feature is ticked in plan/planning.md, which completes the planning epic's open list.

🤖 Generated with Claude Code

Update: I self-merged #157, inspect. Right-click a door, impediment or NPC to see every way past it, each marked available or not with the reason ('Force it: you are Strong'; 'Hack the panel: needs a multitool or Electronics'). The review confirmed it never contradicts what the actions allow. 'Ways past' is ticked.

Building now: digital impediments, traits and door routes as data, loot (gear taken out reaches your inventory once), and inside contact (a contact planted in planning opens a route).

🤖 Generated with Claude Code

Update: I self-merged #158, digital impediments:

  • a door access reader;
  • a camera network;
  • an alarm panel, where the alarm is now a level;
  • turret control;
  • a firewall node in the cyber layer;
  • an air-gapped data store.

They're all on the hacking ladder, with inspect routes. The review found no blockers, and it's at ruleset 12. With #150, the slice's floor of 5 physical and 5 digital impediments is met.

Next: #160, traits and door routes as data, is in review and will be rebased over this. Loot, inside contact and the bot fix for the flaky smoke (#146) are being built.

🤖 Generated with Claude Code

Update: I self-merged #160, the first data-definitions slice.

  • Traits and locked-door routes now come from an embedded core pack (atproto JSON, DAG-CBOR, CID), instead of Rust constants.
  • Missions pin the pack snapshot's CID.
  • The debug menu has Load definitions and Spawn kind here, so a new impediment kind can be added mid-run with no recompile.
  • Inspect reads its routes from the pack.

The review found no blockers, and it's at ruleset 13.

Next to merge: #159 (inside contact), then #161 (loot). The bot and smoke fix for #146 is in progress.

🤖 Generated with Claude Code

Update: I self-merged #162, which fixes the flaky smokes (#146, closed). The smoke bot now opens doors by keycard, by force or through the panel, leaves a held hack map, and passes planning. loop-smoke and hosted-smoke each passed 3 times in a row. It's scripts only.

Building now: hit odds shown before acting, and separate alarm and trace meters. #159 (inside contact) is rebasing, then #161 (loot) follows.

🤖 Generated with Claude Code

Update: I self-merged #159, inside contact. A contact planted during planning waits on floor 1 and opens the nearest locked door or shutter when asked, but may refuse or bolt if alarmed guards are near. The review caught a missing ruleset bump, which is now fixed at ruleset 14. 'Inside contact' is ticked.

Next: #161 (loot) is rebasing now. Hit odds, and separate alarm and trace meters, are being built.

🤖 Generated with Claude Code

Update: I self-merged #163, hit odds. Before you shoot, the target shows 'Shoot guard: N%', computed by the same function that resolves the shot, with the same random draw, so outcomes and replays are unchanged. It only covers visible targets. The review found no blockers, and 'Hit odds' is ticked.

Building now: loot (#161, rebasing), separate alarm and trace meters, and hacking layers in play (encrypted doors).

🤖 Generated with Claude Code

Update: I self-merged #161, loot. Some missions now hold gear worth taking. Getting out with it adds it to your inventory exactly once (a retried settlement can't duplicate it), and a failed run brings nothing home. The settlement and Runner screens show what you got. Ruleset 15. The review found no blockers, and 'Loot' is ticked.

Building now: separate alarm and trace, hacking layers in play (encrypted doors), and melee.

🤖 Generated with Claude Code

Update: I self-merged #165, encrypted layers. From floor 2, some doors and systems carry an encrypted layer that must be decrypted before the hack can unlock them. The hack view shows the layer stack, and traits gate which layers you can try. Ruleset 17. The review's inspect issue was fixed and re-checked, and 'Layers in play' is ticked.

In progress: #164 is being reworked per your note (trapped means Surrender, which ends as Arrested). Melee and the quiet and loud bot policies are also being built.

🤖 Generated with Claude Code

Update: I self-merged #166, quiet and loud bot policies. On seeds 0, 1 and 42, both policies clear floors 1 and 2: the quiet bot never forces anything, and the loud one forces doors, including on floor 2, which has no keycard. It's bot code only, and the core tests pass on the rebased tree (405). 'Two ways through' is ticked.

#164 has been reworked per your note: when you're trapped, you can Surrender, and the mission ends Arrested (no pay, no loot, with a TODO for capture consequences). It's in review now.

🤖 Generated with Claude Code

Update: I self-merged:

  • #164, alarm and trace: separate realspace alarm and net trace meters with thresholds. Alarm can lock the floor down, and per your note, a trapped runner can now Surrender, which ends the mission as Arrested (no pay, no loot, with a TODO for capture consequences). Ruleset 18. The review found one compile fix, which I applied.
  • #121, standing: the shared facts and conditions crate, the first step toward the shop.

In review: #167, melee (a Strong runner with a baton clears floors by melee alone), being rebased. Building: impediments and NPCs as data, and rejecting chrome.

🤖 Generated with Claude Code

Update: I self-merged #167, melee. Bump into an adjacent enemy to strike quietly, with no ammo; the odds show first and come from the same function that resolves the strike. Strong and a baton make it viable: a melee-only bot clears floors 1 and 2. Ruleset 19, reviewed, and 'Melee' is ticked. That completes the combat epic's open list.

Building: impediments and NPCs as data, rejecting chrome, and the shop (stock hidden by standing).

🤖 Generated with Claude Code

Update: I self-merged #169, rejecting chrome.

  • Installing an implant deals a rejection scar in one of your three domains: physiological, psychological or digital (the chrome's own OS).
  • It advances on a real-time clock unless dosed or treated, and treatments depend on the domain.
  • When planning draws it, it becomes a mission complication.
  • The Runner screen shows its stage and when the next dose is due.

Ruleset 21. The review found no blockers, and 'Rejecting chrome' is ticked.

In review: #170 (surrender at any time) and #173 (the shop, with stock hidden by standing). Bug filed and being fixed: #171, the debug menu panics with more than 38 commands. Waiting on you: #168, the plan edit for licences and rejection domains.

🤖 Generated with Claude Code

Update: I self-merged #170: Surrender is always available, as you asked. It asks you to confirm, and says "YOU'RE TRAPPED" when you are; bots never use it. Ruleset 23.

In review or fix: #173 (the shop; its purchase ids are being made per player), #171 (the debug-menu panic), and impediments and NPCs as data. Waiting on you: #168.

🤖 Generated with Claude Code

Update: I self-merged #174, which fixes #171 (closed): the debug menu no longer panics past 38 commands. Each command declares its own optional shortcut, and commands without one are reached by arrows or a click. The mouse smoke passes at the default viewport again.

🤖 Generated with Claude Code

Update: I self-merged #173, the shop.

  • There's a Shop tab with vendors priced in money.
  • A listing and a whole store stay invisible until you have the right standing fact (the debug menu can grant it), and hidden stock never leaks.
  • Purchases go through the ledger into your inventory exactly once.
  • Listings have room for a future licence.
  • The review caught a purchase-id clash between players, which is fixed.

Ruleset 24, and 'Shop' is ticked. Equipping what you buy is next.

In review: #172 (impediments, systems and NPCs as data). Building: more threats (drones and cyber enemies), and New run with seed. Waiting on you: #168.

🤖 Generated with Claude Code

Update: I self-merged #172: impediments, digital systems and NPCs are now data.

  • All of them come from the core pack.
  • A relation store replaces the hard-coded links.
  • NPC menus come from pack routes.
  • A kind that exists only in a debug-loaded pack can be spawned and bypassed.
  • Requirements are opaque ids that protokill's rules check, so the engine crate stays game-agnostic.

Behaviour is unchanged from before. The review's three blockers were fixed and re-checked. Ruleset 25, and the feature is ticked.

Building: more threats (drones and cyber enemies, now on top of the pack), New run with seed, and equipping what you buy. Waiting on you: #168.

🤖 Generated with Claude Code

Update: I self-merged #175, New run with seed… in the debug menu, as you proposed. You type a seed and the local run restarts with it, flagged as debug from turn 0. It's reachable before a session starts. In hosted play the entry is disabled, with the reason given. There are no URL parameters on the player path. Frontend only, and the smokes pass.

🤖 Generated with Claude Code

Update: I self-merged #176, equip what you buy.

  • There's a standing equipped rig, with a Gear section on the strategic screens.
  • Equip and unequip are validated by the rig rules, with plain-English reasons when something won't fit.
  • Planning's loadout defaults to what you have equipped, and bought chrome is installed the same way.

Ruleset 27. The review found no blockers. That closes 'Buy and equip', and with Shop and Loot, the rig epic's economy loop works end to end: buy, equip, take on a job, and bring loot home.

Building: more threats (drones and cyber enemies), hacking agents (heavy versus swarm), and the camera following the runner. Waiting on you: #168.

🤖 Generated with Claude Code

Update: I self-merged #177, more threats, both defined in the data pack and placed from floor 2 up:

  • Drones patrol and are hard to bluff, but you can take them over by hacking (not through a firewall).
  • Cyberware enforcers are tough in melee, and their implants can be hacked or jammed, which raises trace.

Ruleset 28. The review's inspect issue was fixed and re-checked. That closes the NPC epic's open list (guards, staff, interactions, the inside contact and more threats).

A new standing rule in the agent instructions: inspect, menus and hints must use the same check as the action, with a test (reviews caught 4 mismatches today).

Building: hacking agents (heavy versus swarm), the camera following the runner, and items and planning actions as data. Waiting on you: #168.

🤖 Generated with Claude Code

Update: I self-merged #178, the camera follows the runner. It has a 4-cell dead zone, and clicks map correctly after scrolling. The debug menu can generate a seeded 160x64 test floor, and the mouse smoke now covers scrolling and a scrolled click. No ruleset change. I rebased it over the devlog myself.

Building: hacking agents (heavy versus swarm), items and planning actions as data, and the facilities solver with a debug 'Floor report' (the first step of #118's procgen plan). Waiting on you: #168.

🤖 Generated with Claude Code

Update: I self-merged #179, hacking agents. Deck programs (placeholders 'Heavy': strong, one agent; 'Swarm': weak, several and fast) and traits set an agent's strength (the odds shown are the odds used), count (systems held at once, bounded by bandwidth) and speed (holds refreshed per turn). There are no numbers on the runner. Drone takeovers use the same odds. Ruleset 29.

In review or fix: #180 (items and planning actions as data; packs will now reject unknown fields). Building: the facilities solver with a 'Floor report', and contracts built from features (brief, client's view and truth). Waiting on you: #168.

🤖 Generated with Claude Code

Update: I self-merged #180, items and planning actions as data. Rig items, hacking deck programs and planning actions (with their costs and gates) now live in the core pack, and packs reject unknown fields and types at load time. A debug-loaded pack can add a new item or planning action with no Rust change. Behaviour is identical to before. Ruleset 30. That closes the data-definitions epic's content work; server-pushed packs remain.

In review: #181 (facilities solver and Floor report) and #182 (contracts from features). Building: direct and auto hacking agents, Illiterate, and hidden info. Waiting on you: #168.

🤖 Generated with Claude Code

Update: I self-merged #181, the facilities solver.

  • A new game-agnostic protokill_engine_reach crate solves reachability over gated edges.
  • Its gates come from the game's own pack routes, so the solver and the game can't disagree.
  • The generators' completability check now runs on it, with identical floors across 1,200 seeded floors.
  • The debug menu has a Floor report showing what's reachable, and why, for five sample runner types (baseline, quiet, force, net, full).

This is the first step of #118's procgen plan. 'Solver' is ticked.

Next to merge: #182, contracts (rebasing). Building: auto agents, Illiterate, and hidden info.

🤖 Generated with Claude Code

Update: I self-merged #182, contracts from features.

  • Each job is seeded from a goal, hidden facts and complications.
  • It has three views: the brief, the client's view and the truth.
  • The board and planning show only the brief, and the truth stays on the server.
  • Hidden facts appear as data shards on the map.

Ruleset 31. The review confirmed nothing leaks early. 'Contracts from features' is ticked.

Next: 'Hidden info' (reading a shard reveals the truth, and you choose to carry on, bail out or turn on the client) can start now. Building: auto agents and Illiterate.

🤖 Generated with Claude Code

Update: I self-merged #185, hidden info.

  • Reading a data shard reveals the truth behind the brief ('The package is a person').
  • You then choose to carry on, bail out (a new Bailed ending that pays half the bounty, as a placeholder, through the ledger, and leaves the loot behind), or turn on the client (recorded in the outcome; its strategic fallout comes later).
  • Facts reach the browser only once revealed.

Ruleset 34. The review found no blockers. That closes 'Hidden info'.

In review: #183 (Illiterate) and #184 (direct and auto agents). Building: scrolling sections, the strategic scheduler, and signatures.

🤖 Generated with Claude Code

Update: I self-merged #184, direct and auto hacking agents.

  • Direct: drive an agent yourself; your body is pinned while the system is contested.
  • Auto: with an 'Auto mod' deck program, agents hold and defend systems (the Defended rung) while you move in realspace, up to your agent count.
  • Numpad '.' switches modes, and the status line lists your auto holds.

Ruleset 35. The review's flaky tests are now seeded.

Agent instructions gained two rules today: tests always use a fixed seed, and the ruleset is main's plus one at the final rebase.

In fix: #183 (Illiterate). Building: scrolling sections, the strategic scheduler, and signatures.

🤖 Generated with Claude Code

Update: I self-merged #183, Illiterate, a chosen drawback from a lifepath.

  • Terminal text and data shards read as scrambled glyphs, and only in-world text scrambles; game feedback stays readable.
  • Reading routes are closed, and inspect says why.
  • Shards reveal nothing to them.
  • They can still hack through agents and finish by other routes, and the solver has an Illiterate probe kit to check that.

Ruleset 36. All review blockers were fixed and re-checked.

In review: #186 (the deadline scheduler) and #187 (signatures). Building: scrolling sections, the lock-out program, and laser, wired and wireless links. Waiting on you: #168.

🤖 Generated with Claude Code

Update: I self-merged #188, scrolling sections.

  • Floors 3 and up are now 120x48, bigger than the screen, and divided into named sections (Lobby, Offices, Server hall) joined by open doors.
  • The section name shows in the status line and the Floor report.
  • The solver checks every floor (320 checked in review), and a turn takes about 0.8 ms.
  • Floors 1–2 are unchanged. Ruleset 37. 'Scrolling sections' is ticked.

Bug filed: #189, the bot loops in a flatlined hack session on floors 3–4 (pre-existing); a fix is in progress. In fix: #186 (the scheduler; one last clock-read race). Rebasing: #187 (signatures, ruleset 38). Building: lock-out and hacking links.

🤖 Generated with Claude Code

Update: I self-merged #187, signatures. Inspecting an enemy now shows its visible kit and how augmented it is ('light chrome', 'heavily augmented'). Implant detail appears only with a revealing trait (Electronics, Medic) or a tool, and only for enemies in sight. Ruleset 38. The review found no leaks, and 'Signatures' is ticked.

🤖 Generated with Claude Code

Update: I self-merged #186, the strategic scheduler.

  • Deadlines replace the server's world-tick loop.
  • Offer expiry, recovery, accruals, scar stages and mission deadlines fire on time in world time, with downtime cut out.
  • Every action first processes everything due at its own moment, from a single clock read, and a mission settled past its deadline counts as abandoned.
  • Restarts and the audit stay clean.
  • Old tick fields remain as deprecated aliases for now.

Three review rounds caught timing races, all fixed. 'Scheduler' is ticked.

Building: the lock-out program, hacking links, and the bot fix for #189.

🤖 Generated with Claude Code

PR #190 is open and fixes #189 (the bot looping in a flatlined hack session). The bot now clears floors 1–4 on seeds 0, 1 and 42 with both quiet and loud policies, and CI is green. A blocker review is running, and I'll self-merge if it's clean. One choice worth knowing: floors 3 and up still have placeholder combat, so the extended test starts the runner at 30 HP, like the other procedural-floor tests.

I self-merged #190 (bot flatline fix). The review found no blockers and CI passed (456 tests), and #189 is closed. I filed the review's main non-blocking note as a new issue: a shutter that a hack unlocks while the bot can't see it stays blocked in the bot's memory. No fixed seed hits it yet.

Lock-out program is up as PR #192, ruleset 39. You can lock out a held system once its app layer has fallen, if you carry the program and have the trait. A locked-out system starts no new programs and gets no ICE response. Inspect shows a "Lock it out" route, and the hack view and status bar show the state. CI, the wasm tests and the mouse smoke test all pass. An Opus blocker review is running.

Status update:

  • #193 "While you were away" is open. When you come back, the strategic screen lists what changed while you were gone. CI and the smoke tests pass, and a blocker review is running.
  • #194, the fix for #191, is in review.
  • #192 (lock-out) went back to its implementer with four blockers: locked-out cameras and turrets keep attacking, inspect ignores free compute slots, the game deals lock-out cards you can't play, and the trait gate isn't in the pack data.
  • Started: Wounds replace HP (plan/scars, first slice from design/wounds.md). A hit draws a wound card or nothing, thresholds decide stagger, down or dead, and the heart glyph is replaced by a wound row.

Your next action: answer whether a mind should be a room in the hack graph, with social moves running through the existing five layers.

I self-merged #194, the bot shutter fix. The review found no blockers and CI passed (457 tests), and #191 is closed. I filed the review's termination note as #195: a person the bot bluffs out of the way can keep stepping back into its path, and nothing now stops the bot bluffing them forever. No seed hits it yet. Your next action is unchanged: should a mind be a room in the hack graph?

Hacking links (laser, wired and wireless) are up as #196. CI, the wasm tests and the mouse smoke test pass, and the bots still clear their floors. A blocker review is running. #196 and #192 both use ruleset 39, so whichever merges second bumps to 40.

I self-merged #197, the fix for the bot bluffing the same person forever. The review found no blockers and CI passed, and #195 is closed.

Waiting on their implementers:

  • #192 lock-out: 4 blockers.
  • #193 while you were away: an unbounded world-clock skip.
  • #196 hacking links: wired sessions never break, and the air-gapped store can be copied from range.

Wounds replace HP is in progress. Your next action: the daemon questions in chat. Is reach the single dial, and are daemons fitted parts with costs?

Design PR #198 is ready for your review: the layer model extended with minds and daemons (design/equipment.md §3–5, plus one new Daemons feature in plan/hacking.md). Layers now read as "a hack is a climb", minds are components, daemons are kept general, and alerts travel as traffic. Three open questions are in §13. Your next action: review and merge #198.

I self-merged #193, While you were away, and ticked it in plan/strategic. Players who return after more than 60 s see what changed while they were out. The debug time skip is capped at 3 days and refused on hosted servers. Two reviews; CI and the wasm tests pass.

Still in progress: #192 lock-out, #196 hacking links, and wounds.

Your next action: review #198, the design PR. I'll amend it with the hacking terms (op, intrusion, breach, hack) once you've confirmed them in chat.

#198 now has the agreed hacking terms: op, intrusion, breach and hack. It also says that getting into cyberspace is itself an intrusion: creds pass it quietly, or you can intrude on the door controller directly. That adds a new feature to plan/hacking, "Getting in". I started multi-map jobs (plan/contracts): short-sequence contracts with stages, banked results, and a choice to bail between maps. Your next action: review and merge #198, which unblocks the Getting in and Daemons slices.

I cleaned up the stale PRs:

  • #168: rebased. The rejection-domains half had already shipped in #169, so only the Licences line is left.
  • #105: rewritten as a short perks note built on traits (no ranks, no tree, a cost in another currency).
  • #134: self-merged. It's a guard test that keeps store-v4 at schema 4; I checked it against main and clippy passes.
  • #123: being rebased.

Ready for you, all merging cleanly: #198, #168 and #105.

Thanks for merging #198. Getting in and Daemons are now unblocked. I'll dispatch Getting in once #196 (hacking links) lands, because both rewrite the same terminal-connect code and would conflict. Wounds replace HP is up as #199; I'll review it when its agent finishes. Ready for you: #168 and #105.

@permadeath.com, a test mention. Two questions are still open:

  1. Is a mind's voice (speech) a connection into the mind with its own stack, so social moves are a climb like any other intrusion?
  2. Wounds (#199): being down currently means death after 3 turns, because treatment doesn't exist yet. Is that OK for the first slice, as long as surrender stays available?

I self-merged #123, the AI preference record reader. It's the library only, so "Preference reader" stays open in plan/opt-out until the service calls it. The review found no blockers and CI passed.

#203 is ready for you. It moves the open-question sections out of 13 design notes, folds the answered questions into the text, and turns each "(question N)" reference into a stated rule or default. I'll ask you the remaining questions in chat, a few at a time, starting with the ones that affect work in flight.

I self-merged #192, the lock-out deck program, and ticked it in plan/hacking. You can lock out a held system once its app layer has fallen, if you carry the program and have the Electronics trait (gated through pack routes). A locked-out terminal, camera or turret starts no new programs and gets no ICE response. The game only deals lock-out cards when you carry the program. Two reviews; CI, the wasm tests and the mouse smoke test pass, and the bots still clear floors 1–4. Main is now at ruleset 39.

I self-merged #204, multi-map jobs, and ticked it in plan/contracts. Short-sequence contracts have stages: each stage banks its pay, and between maps you continue or bail. Each stage draws a fresh secret seed, and a sequence left idle between stages auto-bails after 1 hour (a placeholder). Two reviews; CI, the wasm tests and the loop smoke test pass. Follow-ups are #205 (freeze banked stage rows, plus tests), and an implementer has started on it.

I self-merged #206, which fixes #205: banked stage rows are now frozen, and there are tests for the partial void, the auto-bail race and the restart backfill. The review found no blockers, and the retry-based tests passed 120 of 120 runs. I also filed and dispatched #207: bots get stuck behind live laser grids on some tutorial layouts.

I self-merged #196, laser, wired and wireless links, and ticked it in plan/hacking. Each link trades reach for exposure: wireless costs +2 trace and laser +1, and every link breaks through the same gate inspect uses. The air-gapped store needs hands on it. Link stats live in the pack. Three reviews; CI, the wasm tests and the mouse smoke test pass. Main is now at ruleset 40. I also started Getting in: entering a terminal takes creds or an intrusion, and door controllers can be intruded directly.

I self-merged #208, big-map view, and ticked it in plan/facilities:

  • free look (Space or a middle drag, with the arrow keys to pan);
  • an overview of explored areas with click-to-travel (O);
  • edge markers for things off screen;
  • section labels.

The review found no blockers, CI passed, and the mouse smoke now covers a floor bigger than the screen. Follow-up: the noise marker should be labelled as a suspicious guard (#209).

Overnight, item 1: the brainstorms you asked for are up for review as #214.

  • 100 job-deck cards and 100 street-deck cards;
  • about 40 planning traits, most with a cost;
  • lifepaths as six stages that alternate rolls and choices, with one-in-a-thousand twists and hard starts;
  • the first twelve enemies, written against the current schemas.

Also:

  • #199 (wounds) passed its re-review apart from a rebase onto big-map, which is in progress. The long bot floors turned out to be an old bot loop that re-hacks terminals, not wound fights; filed as #213.
  • 19 agents are running, at the 20-agent cap.

I self-merged #199, Wounds replace HP, and ticked it in plan/scars.

  • Hits: a hit deals a wound card or nothing, drawn on a per-shot seeded stream. Most hits do nothing.
  • Thresholds: the summed wound aspects decide stagger, down and dead.
  • Down: you can't move, a connecting hit kills you, and you can self-stabilise when no enemy is in sight.
  • Display: a wound row replaces the HP gauge.
  • Systems: cameras and turrets use the same two-step shot.

Two reviews; CI, the wasm tests, and the mouse and effects smoke tests pass. Ruleset 41. Next is the play-doctor microslice, starting with moving the wound rules into crates/game/wounds.

I self-merged #216: the shared microslice kit plus the first microslice, the rig workbench. To try it, run scripts/microslice.sh --all, then scripts/microslice-serve.sh, and open port 4040 over ssh -L. The review found no blockers, and CI, the wasm tests, and the mouse and effects smoke tests pass. The main game renders the same as before.

Scars, job board, lifepath and floor generator are rebasing now and opening their PRs. Hack session, deck lab, glyph lab, the item viewer and play doctor follow.

Found by the review: scripts/develop.sh fails on main on this box, because wasm-opt rejects Rust 1.97's bulk-memory ops. Filed as #221.

Overnight progress. Self-merged, all reviewed, with CI green:

  • #199 Wounds replace HP.
  • #216 the microslice kit plus the rig workbench.
  • #215 and #233 (+#234 cleanup): the threat and grid rules moved out of Bevy.
  • #226 hack sessions no longer share RNG streams.
  • #235 the floor-generator microslice.

Up for your review:

  • #214 brainstorms (200 planning cards, traits, lifepaths, 12 enemies);
  • #219 toolbot note;
  • #220 planning-hand note;
  • #223 mention toasts (auth scope, so your merge, once its backoff fix lands).

In review or fixing: #217 shader layer, #218 card art, #222 typed clickables, #224 in-sight pane, #225 lifepath, #227 hack deck, #228 windowing, #229 job board, #230 smooth scroll, #231 wounds crate, #232 scars, #236 hack session, #238 3D item viewer. Play doctor and the glyph lab are close.

To try the microslices: scripts/microslice.sh --all && scripts/microslice-serve.sh, then port 4040. scripts/develop.sh is broken by wasm-opt on this box; the fix is in progress (#221).

@permadeath.com Ready for your review: #289 design/objectives.md, the objectives system note: majors and minors, open sources (contract, cards, site or faction, traits), mid-mission complications, backing-out costs (cowardice beats death), solver soundness with runner fit telegraphed, and dead-end counting. https://tangled.org/protokill.sh/protokill/pulls/289/latest. Eight questions for you are in chat.

@permadeath.com Ready for your review: #290 design/debrief.md, the mission debrief note: a mission event log, objective/resource/collateral facts, multi-value reputation scorers per audience, value-scaled chaos, smash-things objectives, high-viz runs and taunts. https://tangled.org/protokill.sh/protokill/pulls/290/latest. #289 (objectives) is updated with smash-things. Questions are in chat.

@permadeath.com Ready for your review: #296 design/gear.md, the gear design-space census from our interview: chassis, variants and mods as trait composition; ports, slots and mounts kept separate from network exposure; a soft-counter invariant; identification tiers for loot and enemies; 13 capability families mapped against main. https://tangled.org/protokill.sh/protokill/pulls/296/latest. Questions are in chat.

@permadeath.com Ready for your review: #301 design/setting.md, the setting note from our interview: not Earth, fragmented relational power, the aborted singularity, biopunk, migration, cryptography broken after everyone switched, melancholy with beauty, and a Stand on Zanzibar-style fragment presentation. https://tangled.org/protokill.sh/protokill/pulls/301. Questions are in chat.

@permadeath.com Overnight plan for the architecture review (#308): changes 1 (one gate everywhere), 2 (data-only versioning) and 4 (Bevy-free wire types) are dispatched to Opus agents; 3 (determinism gaps) is folded into the entity-streams fix. Then 5 to 10 as dependencies clear. Each is merged after a full CI run plus a review, and progress is posted here.

Sign up or Sign in to add to the discussion.