Prototype Leader Skills setup, battle disciplines and capture

Connect all fourteen physical skills and private setup to five card-role battle
modifiers, public posture, Harkonnen skilled capture, death return and optional
own-leader revival through controls, legal AI and persisted room actions.
Nine skill effects and combined-mode/publication acceptance remain unfinished.
Captured replacement entitlement and foreign ghola integration stay guarded.

Fresh publisher/designer research supports skilled-capture publicity through
its unique known physical card; this is recorded as a component inference.
Independent review found and fixed Voice/must-use-leader and invalid saved
assignment/offer cases. A pre-existing Discovery privacy assertion confused
card ID prefixes; the corrected deterministic physical-pair scenario was
independently reviewed. The first broad run was 4410/4411; corrected final
verification passes types/lint, all 4411 offline tests, build and 43 HTTP tests.
Final durations: check 149741ms, build 6659ms, integration 28274ms.
Source fingerprint: aef655ac1ebc0eb2f64294e13d3d27337167312cf8ab17568a166ee9c39f6a85.
Private final report: 20260913-leader-skills/checks-final/report.json.

Three seeded prototype games completed with no rejected bot candidates. Browser
setup/card inspection/refresh passed. Another user-reported power outage stopped
the server; backup and necessary restart restored the same private seat. All
292 games at restart and their original application records remain unchanged;
33 isolated HTTP-test rooms bring the current total to 325. No database reset.

Efficiency: one shared setup/lifecycle implementation supports five connected
role effects; selective existing agents handled catalog/UI/tests/research and
independent reviews. One repeated broad check was required by the reproduced
ID-prefix assertion failure. Aggregate goal counter rose from 65529766 to
67313673 through this checkpoint (1783907), including inherited context and
coordination; this is not a billing or savings estimate. Full goal remains active.

docs/LEADER_SKILLS_RUNTIME.md

@@ -0,0 +1,121 @@
+# Leader Skills first functional prototype
+
+13 September 2026. This is **Prototyped**, with **Partial** rules coverage.
+Normal mode starts and publication remain gated. The [source contract](LEADER_SKILLS_RULES.md)
+contains all fourteen physical cards; the [capture interpretation](LEADER_SKILLS_CAPTURE.md)
+records the fresh publisher/designer search requested by the user.
+
+## Connected boundary
+
+Genuine Basic or Advanced setup supports two through six base factions. Starting
+Treachery Cards, including Harkonnen's extra card, precede simultaneous private
+two-card skill offers. Every faction selects one physical skill and an eligible
+own leader, returns the other card to the shuffled deck, and publicly assigns
+the selected card before traitor choices. No reduced or duplicate skill deck is
+substituted. Expansion rosters and other optional modules remain separate work.
+
+Human controls show each permitted card's normal and battle text, select its
+leader, choose battle posture, and handle an optional own-leader revival draw.
+An undrawn revival offer shows no preview; drawing obliges a selection. All four
+AI profiles follow the same private offers and legal actions, prioritizing the
+five connected disciplines and hiding the assigned leader for battle. This
+simple policy is not calibrated skill strategy.
+
+Warmaster, Master of Assassins, Swordmaster of Ginaz, Killer Medic and
+Prana-Bindu Adept apply +1 to another selected leader from a face-up native
+trainer, or +3 to the surviving assigned leader. They use the physical card's
+actual weapon or defense role, including a legally supplied Portable Snooper.
+They do not grant +4 or change printed strength, leader bounty or Zoal's source
+value. Existing traitor, explosion and Stone Burner precedence is preserved.
+Cheap Hero receives the normal +1 through ordinary leader substitution; this
+is a general-rule inference, without a separately located explicit skill FAQ.
+
+A face-up assigned disc cannot enter a plan. Concealing it while selecting
+another leader gives no skill bonus. Later Voice or card use can remove the
+last alternative: the mandatory remaining skilled disc then moves behind the
+shield automatically, without replaying battle powers. A surviving assignment
+returns face up after battle. Death returns its physical card exactly once.
+An own Ghola revival during unfinished battle preparation retains that stage
+through the new private draw and public posture choice.
+
+Harkonnen capture keeps the card attached to its original leader and publicly
+shows the current captor. Only the captive's lower battle effect can apply;
+neither faction receives its normal effect while captured. A surviving used
+captive returns with the same card. Execution returns the card to the deck;
+that unique formerly public card makes this skilled leader's identity inferable.
+Other captures retain the existing privacy behavior. This narrow publicity
+interpretation follows the physical component procedure, not an express
+publisher instruction to announce a name.
+
+## Missing and guarded work
+
+The other nine cards have readable inventory entries but **missing effects**:
+Bureaucrat, Spice Banker, Diplomat, Mentat, Suk Graduate, Rihani Decipherer,
+Sandmaster, Smuggler and Planetologist. They are not completed features.
+
+Captured-card replacement entitlement remains unresolved. The prototype rejects
+own-leader revival while its original skill is captured, before beginning a
+paid revival or consuming the Ghola/special-Karama action. This is an explicit
+unsupported boundary, not a ruling that the physical game forbids that revival.
+Foreign gholas, expansion rosters and combined optional modules remain gated.
+Before enabling them, add response-driven automatic revival continuation
+coverage; the current before/after revival observer does not certify deferred
+Tleilaxu response chains. The other source-contract timing questions also remain
+open. Skill-specific movement, payments, information, rescues and retreats need
+their own connected controls and saved choices.
+
+## Local prototype entry and evidence
+
+Create a new base-faction local lobby, select its seats and mark them ready.
+Keep its browser cookie. Use the existing backup and atomic version-check tool:
+
+```sh
+node --import tsx tools/start-prototype.ts --profile leader-skills \
+  --db PATH_TO_LOCAL_D1_SQLITE --room EIGHT_CHARACTER_CODE \
+  --version CURRENT_LOBBY_VERSION --out /private/new-prototype-checkpoint
+```
+
+Only that fresh named lobby is initialized; every other room and all seats stay
+intact. Refresh to continue through ordinary authenticated actions. No HTTP
+endpoint dispatches the initializer; do not restart or reset the database.
+
+Focused tests cover physical inventory, setup ordering, private views, all five
+bonuses, Voice legality, capture and return, death, revival and JSON restoration.
+SQLite tests use production room actions across restarted stores and concurrent
+selection: one choice commits, the selected card has one owner, the rejected
+card returns once, and private offers remain owner-only. The local prototype
+writer is separately checked for stale/redeal rejection and preserved seats.
+
+Two seeded four-player simulations (20260913 Basic; 20260914 Advanced) completed
+through genuine setup in 265 and 211 accepted actions with no rejected bot
+candidates. They used Easy, Medium, Hard and Brutal, checked physical skills and
+forces, and restored JSON views at fourteen sampled boundaries in total.
+These were first-batch runs before the later capture and Voice review fixes;
+their private report fingerprints the exact files exercised. They do not prove
+all fourteen effects, Harkonnen capture or every interaction was played.
+
+A later Advanced run including Harkonnen (seed 20260915) completed in 360
+accepted actions, four turns and ten sampled JSON restores, with no rejected
+bot candidates. It exercised the current controller/Voice integration. Exact
+captured-card outcomes are proved by the focused capture scenarios, rather
+than inferred from this sample game's completion.
+
+Browser room `REF6YHN7` was created normally with a human Atreides and Medium
+Harkonnen, then initialized through the backed-up local tool. The human inspected
+both effect bands, assigned Master of Assassins to Thufir Hawat, selected a
+private traitor and reached Storm. Refresh retained that assignment, the AI's
+Swordmaster on Feyd-Rautha, the same Snooper hand and the pending Storm dial.
+Visual inspection caught two squeezed columns in the table sidebar; single-column
+cards restore readable text. This was a focused desktop setup/recovery journey,
+not a complete human game or mobile acceptance.
+
+The second reported power outage stopped the development server. All 292 rooms
+were present with a clean SQLite integrity/foreign-key check; the 291 opening
+rooms matched their hashes. A new online backup preceded the necessary server
+start. The same browser seat then restored both skills, Snooper, private Thufir
+traitor, ten spice and its pending Storm decision. No database reset or old
+game restoration was performed.
+
+Checkpoint-wide check/build/HTTP results and the final source fingerprint are
+recorded in the commit and private verification report. Full module and combined
+game acceptance remain unfinished.