In nightmare the merc dies most games, so resurrect_merc runs constantly — and Qual-Kehk
detection was failing 100% of the time (5 timeouts in 5 attempts, 107 hover attempts over
12 games). Each failed hunt costs ~40s and the code retried once, so a dead merc cost
~80s in EVERY game. Game length blew out to 185-250s against a normal ~60s.
GameStats._merc_resurrect_failed did not help: log_start_game resets it, so it only ever
suppressed a second attempt within one game. Nothing carried across games.
The name tag template is degenerate rather than merely stale — every grid-sweep "hit"
reported the identical score at unrelated positions:
found name tag at (255, 227) (score 0.424)
found name tag at (1110, 100) (score 0.424)
found name tag at (930, 310) (score 0.424)
so "found" is meaningless; it is matching uniform background.
Fix is cost containment, not detection: a cross-game circuit breaker on GameStats that
log_start_game deliberately does NOT reset — _merc_resurrect_fail_streak and
_merc_resurrect_skip_until, with Bot._MERC_RESURRECT_FAIL_LIMIT=2 and
_MERC_RESURRECT_SKIP_GAMES=15. The retry is also skipped once the streak is >=1, since
that is a second guaranteed-futile 40s hunt. Both counters clear on any successful
resurrect so a transient failure cannot permanently disable resurrecting.
Simulated over 30 games with an undetectable NPC: 60 hunts -> 4 (~40 min -> ~2.7 min).
Measured live: the breaker engaged on game 2 and game times went 250s / 185s -> 14s, 14s,
43s, 71s, 111s.
Still open: recapturing qual_name_tag_white.png is the actual fix for detection.
Co-Authored-By: Claude Opus 5 <[email protected]>
- pickit: run-scoped loot accumulator (item names of every successful
pickup), consumed once per run.
- bot._run_wrapper: at run end (success, battle-fail, and approach-fail
paths) logs "Loot from run_diablo: JAH RUNE, 2x FLAWLESS SAPPHIRE" (or
"nothing picked up") at INFO level.
- game_stats: run_finished events now carry a counted loot dict, so
events_*.jsonl can be mined for per-boss drop value over time.
Co-Authored-By: Claude Fable 5 <[email protected]>
Mirrors the D2R settings check: on the first on_maintenance of a bot
session, validate_build_skill_icons presses each configured skill hotkey
and verifies the icon on the right slot, warning loudly (with a pointer to
tools/set_binds_from_params.py) when a bind does not match params.ini.
Non-blocking; ~5s once per session.
Co-Authored-By: Claude Fable 5 <[email protected]>
Session fixes (2026-06-11), all live-tested over multiple farm games:
- input_layer/hotkey.py: no-arg keyboard.wait() returned on ANY keypress,
silently killing the process right after F11 (all bot threads are
daemons). Now blocks forever like the original keyboard lib. Poll loop
is edge-triggered (no ~20ms refire while a hotkey is held) and callback
exceptions print instead of being swallowed. (Bug 13)
- d2r_image/ocr.py: pytesseract never read PYTESSERACT_TESSERACT_CMD;
tesseract_cmd is now wired from env var / PATH / winget default, fixing
exit 0xC0000135 on every OCR call. (Bug 14)
- bnip/utils.py + config/default.bnip: undefined NipSyntaxError ->
BNipSyntaxError, and Shaefershammer -> Schaefershammer typo; 474 pickit
expressions load (was 473 + parser error). (Bug 15)
- town_manager/a5/bot: A5 WP death-loop containment - per-game WP failure
budget (2 strikes), quick=True direct-path-only retries, sweep trimmed
10->6 steps with 4s select timeouts, A5 select thresholds lowered
(WP 0.62, stash 0.60/0.45; safe - every select is success_func-gated).
Worst case dropped from 10+ min wandering to ~4.5 min contained fail
with fresh-game recovery. (Bug 16)
- bot.py: vendor trip gating - the failure-prone A5->A4 Jamella round
trip now only runs when consumables are needed or 3+ sell items pend.
- town_manager.py: Cain identify skips acts whose Cain timed out this
session (straight to working A5 fallback).
- char/i_char.py: Battle Command buff check waits 0.6-0.8s after the
hotkey (icon fade-in) to reduce double weapon-swaps.
- char/paladin/hammerdin.py: kill_diablo fights with Concentration
instead of Conviction (useless for magic-damage hammers) and drops
mid-fight Redemption downtime - faster kills, merc survives. (Bug 17)
- inventory: stash supports all 6 pages (personal + 5 shared, D2R 2.7+);
gold deposits navigate via OCR-verified select_stash_page instead of
raw 4-tab clicks; rotation %6, shared-first starts at page 5. (Bug 18)
Docs: CLAUDE.md Bugs 13-18; .hermes/plans/dia_run_test_state.md has the
full test log (two complete Diablo kills verified end-to-end today).
Co-Authored-By: Claude Fable 5 <[email protected]>
CRITICAL loot-loss bug: buy_consumables failure called end_game and
returned BEFORE the stash step, so Pindle runes/items piled in inventory
and were never banked (zero stash events across a whole session). Buying
pots is optional (belt refills from drops); stashing loot is the point.
Now warn + re-anchor + fall through to stash. Also raise
max_maintenance_time_s 120->240 so the A5 vendor thrash completes and
reaches stash instead of timing out first (transitional — disappears once
a Diablo run shifts spawns to A4).
Co-Authored-By: Claude Fable 5 <[email protected]>
Malah wanders and her body templates predate the patch; the hunt fails
most games and the failed-hunt position strands the char. Jamella is
static and detects at 0.99+. Travel cost ~30s vs a lost game.
Co-Authored-By: Claude Fable 5 <[email protected]>
- town_manager.identify: return the reached A5 location when Cain fails
after successful travel (returning False made bot.py fall back to the
pre-travel act and run wrong-act pathing all maintenance long)
- town_manager.resurrect: leave last_known_loc breadcrumb when the A4
revive fails after travel; bot.py consumes it to re-anchor
- bot.py: alternate vendor retry prefers A4 Jamella (static, reliable)
over A1; Malah wanders and is the main A5 vendor failure
Co-Authored-By: Claude Fable 5 <[email protected]>
First encounter at char select blindly saved whichever character was
highlighted, locking in a wrong character for the whole session. Now,
when char_name is configured, select it via OCR first and warn if the
saved template name does not match.
Co-Authored-By: Claude Fable 5 <[email protected]>
Root causes from the 2026-06-09 session (51 games, 45 failed):
- Act desync: add TownManager.detect_current_act() and
Bot._verify_town_location(); every maintenance/end-run retry and
fallback now verifies the physical act instead of hardcoding town
starts. open_wp/go_to_act self-heal act mismatches. Never run A1
pathing when travel to A1 failed.
- NPC dialogue: poll action buttons up to 2.5s after click instead of
a single-frame check (premature retry click was closing the dialog).
- Stats integrity: log_end_game skips duplicate calls (phantom 0s
"successful" games were resetting the consecutive-fail breaker);
clear stale failure reason at game start; set chicken flag before
bot.stop() so chickens are no longer labeled "Bot stopped".
- Repair: prefer in-act Larzuk over cross-act Halbu trip (Halbu
detection failed 100% last session and desynced the act state).
Documented as Bugs 9-12 in CLAUDE.md.
Co-Authored-By: Claude Fable 5 <[email protected]>
Reliability
- All run approach() methods now set approach_fail_step before every return False;
bot.py _run_wrapper reads it and includes [step: X] in Discord/log failure reasons
- on_maintenance() sets _maintenance_step before each town step (heal, identify,
buy_consumables, stash_items, repair, resurrect_merc, gamble) for the same coverage
- on_init() logs startup line: char=X | difficulty=Y | routes=[...] each game
Per-run disable (bot runs as long as possible)
- GameStats tracks per-run failure counts across Bot instances (previously reset each game)
- After disable_run_after_failures consecutive failures a run is disabled for the session;
game-level consecutive-fail counter resets so the bot continues on remaining routes
- game_controller no longer quits on max_consecutive_fails if active routes remain;
only exits when all routes are disabled
Maintenance timeout (params.ini: max_maintenance_time_s=120)
- Hard 120 s wall on the entire town maintenance loop; checked between each major step
and before every retry (buy_consumables, stash, repair, resurrect_merc, gamble)
- On timeout: error screenshot + Discord, then trigger end_game → save-and-exit → rejoin
Bug fixes
- Win11 mouse overshoot: mouse_move() uses SetCursorPos + zero-delta MOUSEEVENTF_MOVE
- _curr_loc = True propagation: TownManager.identify() now returns the act Location enum
- DAMAGED KeyError in pickit: added ItemQualityKeyword.Damaged.value to NTIP_ALIAS_QUALITY_MAP
- A4 WP interaction range: force-move character to WP stone before select_by_template
- NPC click blocked by equipped-area guard: open_npc_menu() closes inventory if open
Docs
- Added CLAUDE.md: AI working guide with step tables, bug history, debugging tips
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
open_npc_menu timeout reduced 35s → 20s and now saves a screenshot when
it gives up. buy_consumables and stash failures after their retry now
trigger end_game (save & exit) instead of silently continuing with a bad
location, preventing the bot from running the next game in a broken state.
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
When restart_d2r_when_stuck is enabled, spawn a fresh Python process
(same main.py) and exit immediately rather than killing and relaunching
D2R. The new bot detects D2R is already running and skips launching it,
preserving the game session. D2R is only killed on deliberate exits
(safe_exit) and the initial auto_login launch.
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
Adds Messenger.send_error across the Discord (red embed with screenshot attached)
and generic (text-only) APIs. Bot._save_error_screenshot now also pushes the
failure to the configured messenger after saving the screenshot to disk, so each
approach/battle/exception failure is reviewable in Discord with the visual.
Gated by new config discord_log_errors ([general], default 1) and the
[discord_events] error toggle. Both optional/backward-compatible. Verified wiring
and the suppression path via a stubbed messenger.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
When every run has been auto-disabled there are no routes left to run, so save a
session report and shut down via safe_exit() instead of calling restart_or_exit
(which would needlessly restart D2R into empty games when restart_d2r_when_stuck
is enabled).
Co-Authored-By: Claude Opus 4.8 <[email protected]>
Recovery: each run now has its own consecutive-failure counter. After
disable_run_after_failures (default 5) consecutive failures, that single run is
disabled for the rest of the session and the bot keeps doing the other runs
instead of stopping. A success resets the counter. If every run is disabled the
bot stops for investigation. Disable is in-memory only (restart re-enables).
Diagnostics: on every run failure (approach, battle, or exception) the bot saves
a timestamped screenshot to log/screenshots/error/ named with the run, reason,
and game/run counters so logs and visuals can be cross-referenced. Gated by the
new error_screenshots config (falls back to info_screenshots). The error/ dir is
routed through log rotation.
Adds config keys error_screenshots and disable_run_after_failures (both optional,
backward compatible) and docs/recovery_and_error_logging.md. Verified bot startup,
config parsing, Bot construction, the disable/reset logic, and screenshot writing.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
- Implementation of centralized cooperative shutdown mechanism (C10) to replace dangerous thread killing.
- Consolidation of all timing signatures through jittered wait() for anti-cheat stealth (C1).
- Enhanced HealthManager reactivity with smart mana potion fallback and premature panel-closing protection.
- Refactored waypoint stealth logic to simulate human-like mis-aim instead of area-breaking misclicks.
- Implementation of stateless 'garbage item' filtering in PickIt to eliminate ghost item loops and OCR artifacts.
- Optimization of inventory management to automatically stash protected items that cannot be sold.
- Suppression of individual game failure notifications on Discord; alerts now trigger on 5+ consecutive fails.
- i_char.py: rewrite _pre_buff_cta() to verify each weapon switch via BC
skill-bar template; detects wrong slot on game start (leftover from
interrupted buff cycle), retries failed switches once, logs clearly
when stuck on CTA slot to prevent dying with wrong weapon in combat
- personal.py: protect needed consumables from sell/drop in inspect_items;
check get_needs() before marking a pot for discard — if the belt needs
that pot type, skip it so fill_up_belt_from_inventory can restock later
- bot.py: add fill_up_belt_from_inventory + update_pot_needs after
buy_consumables so inventory pots reach the belt even when out of gold;
add town-heal loop at start of on_maintenance to drink health/rejuv pots
until HP >= 95% before the next run (health manager is paused in town)
- game_stats.py: add rolling 10-game pickup health check; warns in log and
sends Discord alert when zero item pickups occur across 10 games while
chickens or merc deaths are present
- params.ini: fix show_belt=n -> show_belt=k (belt key was wrong, causing
1.5s wasted recovery attempt every first game in a session)
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
- Updated bot behavior to end the game when no TP charges are left, instead of attempting to walk back to town.
- Refined item processing logic in `processing_helpers.py` to use identity checks for item comparisons and improved item removal methods.
- Enhanced game stats logging to include failure reasons and send Discord notifications for failed games.
- Improved health management logic to trigger chicken behavior based on health percentage after rapid rejuvination.
- Added error handling for item pickup failures in `pickit.py` to blacklist items that cannot be picked up.
- Introduced walking fallback nodes for Countess, Andariel, Mephisto, and Baal runs in `pather.py`.
- Implemented path data validation in run scripts for Countess, Andariel, Mephisto, and Baal to ensure paths are correctly defined.
- Enhanced item identification and stash processes in `diablo.py` with retry logic for failures.
- Updated town management to handle identification failures gracefully and log appropriate warnings.
- Improved character selection logic to provide clearer logging messages.
- Modified game restart logic to launch the game via PowerShell for better session handling.
- Added documentation for recording paths for the Countess run and created a PowerShell script for scheduled task management.
- game_controller: set reason on max_game_length, chicken, force_stop
- bot.py on_end_game: set default reason when failed=True has no reason yet
- bot.py save_and_exit failure: set reason before stop()
- bot.py restart_or_exit: capture error message as failure reason
- All retry paths (stash, buy_consumables, heal, repair) now use known-good
location constants instead of corrupted self._curr_loc (False)
- Guard gamble->stash chain against False location
- Guard wait_for_tp against False location
- Add run_trav to route order (travincal then pindle)
- Fix merc icon detection: expanded ROI from 40x40 to 60x60 (matchTemplate requires search area > template)
- Add MercPanelText screen object using mercenary_screen.png for reliable merc alive check
- Bot merc check now uses MercPanelText instead of generic LeftPanel
- Diablo: wait for boss to spawn before attacking (up to 15s), use conviction aura
- Diablo/Pindle attack lengths increased (diablo 3->10s, pindle 3->8s)
- Aura key switch delay increased from 50-100ms to 200-300ms for reliable activation
- Add conviction hotkey (F5) to hammerdin config
- Move Pindle CTA pre-buff from town to after entering red portal
- More aggressive healing thresholds (health potion at 98%, rejuv at 85%)
- Max game length increased from 160s to 600s for full Diablo runs
- Crop A5_RED_PORTAL template to remove empty space (center now aligns with portal)
- Fix merc_icon ROI from 10,9,50,50 to 15,14,40,40 (matches 40x40 merc templates)
- Lower MercIcon threshold from 0.9 to 0.85, add MERC_A2_2 variant
- Add merc panel confirmation with 'O' key before resurrect (prevents false resurrect when icon detection fails)
- Override capabilities to can_teleport_natively
- src/utils/misc.py: move_d2r_window now calculates outer window position
to place client area at target coords (was incorrectly setting outer
window to target, shifting client area unpredictably)
- src/bot.py: Call move_d2r_window(5,98) at game start to restore original
working client area position and prevent 13px offset drift
- src/screen.py: Add force param to find_and_set_window_position/set_window_position
to bypass early-return guard and skip wait when called programmatically
- src/run/pindle.py: Force window re-detection before each traverse_nodes and
select_by_template call to handle offset drift between runs
- src/pather.py: Add fallback full-image search with lower threshold (0.55) when
cropped ROI search fails in find_abs_node_pos
- src/bot.py: Move D2R window to stable position (0,0) at game start via
move_d2r_window; skip merc resurrect after first failure (no gold)
- src/game_stats.py: Add _merc_resurrect_failed flag (reset each game)
- src/utils/misc.py: Add move_d2r_window() function
- config/params.ini: Updated teleport=b, show_belt=k, override_capabilities,
restore_settings_from_backup_key=insert, graphic_debugger_key=delete
- Add utils/log_rotation.py: safe_imwrite() auto-rotates screenshot dirs
when they exceed configured file count or size limits
- Add [log_rotation] config section in params.ini (pickit/info/items
max_files, max_mb, discord_notify_rotation)
- Replace 40 cv2.imwrite() calls with safe_imwrite() across 13 files
- Add failure reason tracking: bot.py catches run exceptions, stores
reason in game_stats, Discord message includes the error
- Add Discord notification when log rotation deletes old files
- Prevents disk-full crashes that stopped botty on May 30
- Read character .key file from Saved Games / D2R install dir / APPDATA
- Auto-fill empty hotkeys in [char] section (inventory, belt, potions, etc.)
- Match skill slots to configured skills for validation
- Normalize key variants (left alt ~ alt, left shift ~ shift)
- params.ini values always take priority over detected bindings
- Warn on genuine mismatches between config and .key file
- on_end_run: calls maybe_afk_break() after returning to town
- run selection loop: checks should_skip_run() before each run
- on_init: reshuffles _do_runs_reset if reshuffle_each_rotation is enabled
Also scaffolded 4 new run stubs: baal, mephisto, andariel, countess
Updates the prebuffs for paladin to use holy shield before entering Pindleskin's portal. Also, adds additional pickit logic for picking up and storing all rare items when uncommenting the sections.
