Basis -- porting notes

Source: basis-20.tar.gz, recovered directly from the still-live ftp.lysator.liu.se LPMud archive mirror (https://ftp.lysator.liu.se/pub/lpmud/mudlibs/MudOS/basis-20.tar.gz, Last-Modified: 05 Aug 2010, itself an untouched 2010-era copy of the original file). No Wayback Machine recovery was actually needed -- the CDX API had no capture of the file itself (only of its parent directory listing), but the live mirror still serves the exact file. Slug basis, number 959, port 40261. 233 files, real adm/obj/master.lpc tree, mudlib root at the archive's top-level basis/ directory.

Author: John Garnett ("Truilkan"), same person as the already-onboarded tmi2 (TMI-2), but this is a distinct, earlier (1993), and much smaller/thinner project -- explicitly described by its own author as "an experiment," "far from a finished polished mudlib," intended "for experimenters who actually want to start with something smaller and less complete than TMI-2." Basis's own std/access/emoted conventions were later inherited by TMI-2 (confirmed by the archive's own README).

1. Extraction and conversion

Pure ASCII/English archive, 0 lossy conversions (as expected -- convert_lib.sh's encoding pass was a no-op). 233 files renamed .c->.lpc, 43 literal .c" references fixed, 3 local angle-bracket includes converted to quotes, 33 files static->nosave.

2. Proactive on-sight checklist findings

• status used as a declared type (AGENTS.md §6.8) in two places, one of which my initial targeted grep for ^\s*status\s / [(,]\s*status\s MISSED: std/i/pager.lpc's nosave private status use_get_char; (the status keyword sits mid-declaration after nosave private, not at line-start or inside a parameter list). Lesson for future archives: the safe general grep is \bstatus\b across the whole tree, filtering out plain-English prose/path-string false positives by hand, not the two narrower patterns AGENTS.md currently documents. Fixed: int.
• new used as an ordinary parameter/local name in two files (adm/obj/simul_efun/resolv_path.lpc, .../replace_string.lpc) -- new(...)/new(class ...) is a hard-reserved keyword on this driver (object-instantiation syntax), unconditionally, unlike the compile-flag-gated ref reserved word (AGENTS.md §6.7). This is a new, previously-uncatalogued reserved-identifier collision class (added as a new AGENTS.md §6.2 entry this session, alongside the sibling class collision below).
• class used as an ordinary parameter name (std/i.lpc's receive_message(string class, string msg)) -- also a hard-reserved keyword (typed-class/struct syntax), same failure shape as new.
• The proactive checklist found nothing else unusual: no ed_start/ ed_cmd, no efun::set/query/delete, no switch...default: empty clause, no hardcoded ports, no this_player() inside the security daemon, no suspicious destruct calls in master.

3. Compile-time driver-compat fixes

• Missing global include file config key. The original archive's own config commented this out ("This is an optional define, and currently doesn't work" -- true on old MudOS, not true here): with the key entirely absent, this driver's config parser still zeroes the destination buffer, sees an empty (thus unquoted-looking) value, and unconditionally pushes a phantom #include "" before literally every compiled file's first line -- error: Cannot read #include file on every single compile, though non-fatal by itself (the rest of the file still compiles). Fixed by adding global include file : <config.h> (a pattern already used by 9 other libs in this corpus for a similar config.h-flavored header). Cross-checked: EVERY other lib's config.fluffos in this corpus sets this key -- worth treating its absence as a red flag on any new archive, not just an omission to fix reactively.
• extract() and log_file() were never real efuns on this driver (classic MudOS efuns, both absent from FluffOS's spec). extract(str, start[, end]) (substring by inclusive index range, 2- and 3-arg forms both used in this archive) and log_file(file, str) (append to a file under the mudlib's own LOG_DIR) both reimplemented as simul_efuns, added to adm/obj/simul_efun.lpc's include chain. See AGENTS.md §6.2's new entries.
• move_object() 2-arg vs 1-arg dialect gap (AGENTS.md §7.158's documented class): std/object/move.lpc's move() called efun::move_object(this_object(), ob) (old MudOS 2-arg form); this driver's move_object() efun takes exactly one argument (the destination) and always moves this_object(). Since the old first argument was always this_object() anyway here, the fix is the simplest case of this class: drop the first argument entirely.
• A vestigial private inherit created an ambiguous duplicate of a needed public function -- a new bug class, not just AGENTS.md §7.48's "private declared in one file, called from an inheriting program" shape. std/living/envars.lpc did private nosave inherit CORE; purely as leftover cruft -- it never actually called any of CORE's query()/set()/delete()/add() methods (it manages its own independent evars mapping). std/living.lpc inherits BOTH std/object/base.lpc (a normal, PUBLIC inherit CORE) and envars.lpc -- giving living.lpc two separate inherited copies of query(), one public (initialized, the one actually wanted) and one private (via the now-removed inherit, its own attrs never initialized since envars.lpc's own create() override shadows CORE's). This driver picked the PRIVATE copy for living.lpc's own unqualified query(a_name)/query(a_permissions) calls, hard-failing with "Illegal to call inherited private function 'query'" -- blocking the ENTIRE player-body compile (std/i, std/admin, std/maker, std/user all failed transitively). Diagnosed by noticing envars.lpc never actually uses anything from its own inherit CORE line. Fixed by deleting the vestigial inherit outright (not by loosening private to protected, which would have "fixed" the compile error but left the SAME wrong, uninitialized copy winning at runtime -- a silent wrong-mapping bug instead of a clean compile error). New AGENTS.md §7 candidate: when two inherited copies of the same base class collide (one public+wanted, one private+vestigial), check whether the private one is actually load-bearing before just relaxing its visibility.
• The most severe bug: a runtime-only closure/function-pointer semantics gap, not a compile error at all. std/object/base.lpc's setup_efun_attributes() builds (: SIMUL_EFUN_OB, "query_efuns" :) -- old MudOS's "call_other-bound function pointer" idiom, meant to redirect query(a_super)/query(a_contains)/query(a_uid)/ query(a_ip_address) through a simul_efun (SIMUL_EFUN_OB is the path string "/adm/obj/simul_efun"). This driver's grammar (per ~/src/fluffos/docs/lpc/types/function.md, "the third type is the call_other function pointer... (: object, function :)") documents this working for ANY expression in the head slot including a bare string, but the ACTUAL implemented grammar (compiler/internal/grammar.y's L_FUNCTION_OPEN L_DEFINED_NAME ',' arg_list ':' ')' production) only builds that special call_other form when the token immediately after (: is a bare, compile-time-resolved NAME followed directly by a comma. A string constant (or even a called this_object()) instead falls through to the generic L_FUNCTION_OPEN comma_expr ':' ')' "expression functional" rule, which just evaluates the WHOLE thing as an ordinary C-style comma-expression and returns the value of the LAST operand, discarding everything before it and calling nothing at all. Confirmed empirically via a temporary catch() + sprintf("%O", ...) probe inserted into query(): invoking the closure printed attr=(: <code>() :) (an expression-functional, not a call_other pointer) and returned the literal string "query_efuns" verbatim -- no call ever happened. That bogus string then got treated as an object/path by the caller (look.lpc's act_ob->query(a_super)), and the driver's own call_other-of-a-string fallback tried compile_object("/query_efuns"), which virtuald.lpc's own generic-virtual-object compile_object() couldn't resolve either, falling back to /adm/obj/virtual/0, a path that ALSO doesn't exist -- triggering the driver's own master::compile_object() -> virtuald::compile_object() -> master::compile_object() self-referential recursion (since master unconditionally re-delegates every failed load back to virtuald), hard-crashing with *Inherit chain too deep: > 30 on literally every look (and anything else touching query(a_super)/query(a_contains)). Fix: use call_other itself as the bare NAME (satisfying the grammar's special-cased shape) with the real target object/path and method name as BOUND arguments -- (: call_other, SIMUL_EFUN_OB, "query_efuns" :). The query()-time argument (key) gets appended after the bound ones at invocation, giving exactly call_other(SIMUL_EFUN_OB, "query_efuns", key), the old semantics. Applied the identical fix to obj/weapon.lpc's (: this_object, "query_damage_string" :) (which ALSO first hit a separate, more obvious compile-time symptom: bare this_object -- no parens -- inside a closure literal is parsed as "call the this_object() EFUN with the following args," a hard compile error distinct from this runtime gap; even fixing that to this_object() -- called -- still only produces a broken expression-functional, not a working call_other pointer, until also rewritten through call_other). This is a new, previously-uncatalogued AGENTS.md §6.2 bug class -- added this session. Given how fundamental and silent this failure mode is (compiles clean, no error until the wrong return value is actually consumed, and even then no crash until something tries to use the bogus value as an object), any future archive using the classic (: obj_or_path_expr, "method" :) idiom for anything other than a bare bound NAME should be checked for this specifically.
• A valid_write()/valid_read() save-file check forgot the driver appends the save extension -- exactly AGENTS.md §7.49's documented class. adm/std/master/valid.lpc's valid_write/valid_read both ended with return (file == data_file(user)); where data_file() (a simul_efun) returns the bare path with no .o extension, while the driver's own save_object()/restore_object() pass the FULL suffixed filename. Every save_data() call -- including the very FIRST one, fired automatically during brand-new character creation by new_user.lpc's setup() -- hard-errored with "Denied write permission in save_object()", silently aborting setup() mid-flow BEFORE it ever reached switch_new_obj(). Symptom: registration looked like it completed (name/password prompts all worked, reached a bare > prompt), but every single typed command produced only the generic fail message, because the connection was still attached to the ORIGINAL login object, never migrated to a real player body at all. Fixed with the standard widening: file == data_file(user) || file == data_file(user) + ".o".
• explode("/", "/") returns an EMPTY array on this SANE_EXPLODE_STRING-built driver, not the two-element ({"", ""}) old MudOS returned (AGENTS.md §6.7's documented class, previously found on tmi2). Hit THREE times in this archive (not just the usual single master::domain_file() boot-time probe): adm/obj/simul_efun/domain_file.lpc, creator_file.lpc, and author_file.lpc all did the same path = explode(str, "/"); if (!path) return 0; switch (path[0]) {...} shape, all three called transitively via master.lpc's own domain_file()/creator_file()/ author_file() delegators, all three hard-erroring "Array index out of bounds" on the driver's own first boot-time probe of "/". Fixed all three with the standard !path || !sizeof(path) widening.
• indices() is not a real efun on this driver -- old MudOS's name for what this driver calls keys(). One hit: bin/user/shell/nickname.lpc.
• privp() is not a real efun on this driver (old MudOS predicate: "does this object have an effective uid, i.e. is it privileged"). wizardp() DOES exist here and was already used correctly elsewhere in the archive; only privp() needed reimplementing (as a simul_efun: geteuid(ob) && strlen(geteuid(ob))). One call site: bin/maker/objects/origin.lpc.
• cat()/tail() are not real efuns on this driver -- classic MudOS file-display efuns (cat(file[, start, num]) writes a file or a line-range of it directly to the current player and returns the line count; tail(file) writes the last 10 lines). Reimplemented as simul_efuns (AGENTS.md §6.2 already flags tail() as a known class; cat() is new). 6 call sites across bin/maker/file/{cat,less,tail}.lpc, bin/maker/objects/log.lpc, adm/std/login/misc.lpc (the login banner!), adm/obj/mailer.lpc.
• MUDOS_VERSION/MUDOS_ARCH are not predefined on this driver -- old MudOS injected these as compile-time macros; this driver predefines __VERSION__/__ARCH__ (double-underscore) instead. Bridged with two #defines in include/config.h.
• MOVE_NO_ROOM used as a switch/case label, but never #defined anywhere (bin/user/objects/give.lpc) -- a genuine pre-existing author typo (AGENTS.md §6.6's class), not a conversion artifact. include/move.h defines MOVE_OK/MOVE_FAILED/ MOVE_ATTACHED/MOVE_FULL/MOVE_TOO_HEAVY/MOVE_TOO_BIG but no MOVE_NO_ROOM. The adjacent fail message ("can't carry any more") matches MOVE_FULL exactly; fixed to that constant.
• A comment-continuation typo, same shape as AGENTS.md §4.1's qhxajh finding but in a header, not a heredoc: adm/daemon/parsed.lpc's doc-comment block has one line missing its // prefix (| direct_object_phrase preposition indirect_object_phrase, meant as a continuation of the // input: line above it), leaving a bare | that's a hard syntax error. Genuine 1992-era author typo, not conversion damage (confirmed against the raw archive bytes). Fixed by adding the missing //.
• **Old MudOS silently allowed rebinding a typed string parameter to hold a mixed * value; this driver enforces the declared type strictly.** bin/maker/objects/call.lpc's do_command(string arg) reused its own arg parameter to hold (mixed *)previous_object()->parse_args(args) + dummy (an array), a genuine 1990-era author habit (untyped/loosely-typed classic LPC), not a conversion bug. Fixed by introducing a separate, correctly mixed *-typed local (parsed_args) instead of reusing arg.
• 19 files' inherit "/bin/bin_m" (some with, some without the leading slash) pointed at a file that no longer exists at that path -- but the content DOES exist, just relocated to /adm/std/security/bin.lpc (that file's own header comment still literally says // file: /std/bin/bin_m.c, confirming the move) without every reference being updated. This is a real, fixable archive-internal path-drift bug (the content exists), distinct from genuinely missing content -- confirmed by checking the moved file's actual implementation matches what every caller expects (a do_command/execute/permission-check base class for bin/ commands). Fixed all 19 inherit statements to point at /adm/std/security/bin. This one bug alone accounted for the majority of the ~60 lpcc_check.sh failures found before any of the other fixes above (Inherit chain too deep cascading through the driver's virtual-object-compile fallback, same self-referential mechanism as the closure bug above, triggered by trying to load a file that plain doesn't exist at the referenced path).
• adm/daemon/whatisd.lpc inherited /bin/daemon/mand, a path that never existed anywhere in the archive; the real file is /adm/daemon/mand.lpc. A simple stale/typo'd path (bin/daemon vs. adm/daemon), fixed directly.

4. Compile-sweep triage: known-acceptable remaining failures (11 of 237)

All confirmed either (a) fragment-only files meant to compile ONLY as part of adm/obj/simul_efun.lpc's monolithic #include chain (existence.lpc, communications.lpc, to_object.lpc, writef.lpc -- all reference macros/globals that only exist once pulled in via the parent file's own earlier includes; the parent file itself compiles and loads cleanly, confirmed live), or (b) genuinely unreachable dead/incomplete original content, none inherited by anything live and none reachable from any in-game command:

• std/i/tsh.lpc -- an abandoned alternate shell implementation (its own header comment: "Brutally hacked up and destroyed by Buddha to install something 'better'"), #includes two headers (commands.h, tsh.h) that never existed anywhere in the archive. Never inherited by anything. bin/maker/shell/tsh.lpc (the tsh command wrapper) calls pobj->tsh(str), which would fail at runtime since no player body inherits this file -- a genuinely broken, pre-existing, never-wired-up command, left as-is.
• std/i/history.lpc -- an incomplete bash-history-recall feature (!!/!N recall syntax) referencing ptr/max/cmd_num variables that are never declared anywhere in the file nor exposed by the QUEUE_ADT it uses. Never inherited by std/i/shell.lpc or anything else live -- genuinely incomplete original content.
• bin/maker/test/ws.lpc -- literally malformed at the type-declaration level (nosave int\nstring do_command(path) -- two conflicting return-type declarations on consecutive lines, plus an untyped K&R- style parameter), under a directory named test/. Never referenced elsewhere. Left as the author's own experimental scratch file.
• bin/maker/objects/clean.lpc -- inherit CLEAN_UP; where CLEAN_UP is never #defined anywhere in the archive (unlike bin_m, there is no relocated file to point this at instead). Never referenced elsewhere.
• bin/maker/status/sockinfo.lpc / opcprof.lpc -- call dump_socket_status() / opcprof(), MudOS-driver-level debug introspection efuns with no FluffOS equivalent at all (not reimplementable in pure LPC). Out of scope, same class as AGENTS.md §7.51's NTOS-heartbeat gap.
• obj/terminal.lpc -- create() calls getuid(this_player()); a standalone lpcc_check.sh isolated-load probe has no interactive this_player(), so this fails ONLY in isolation (confirmed: not preloaded, not cloned anywhere at boot, only ever created via a live player's own action, at which point this_player() is always valid) -- a known lpcc-vs-real-boot false positive (AGENTS.md §10.4).

5. Live playthrough (native driver)

Full flow verified in one continuous session, port 40261: fresh registration (fluffos / Mud@2026) reaching /room/start -> look (correct room description) -> go east (moves to the second room, correct room-broadcast message to the departed room) -> look again (correct new-room description) -> quit (clean disconnect message). Re-login (the restore/existing-password path) verified separately: password accepted, look shows /room/start again (player position is not part of the minimal save data this archive implements -- confirmed intentional/pre-existing scope, not a regression, since core::attrs/save_data() never captured environment in the first place). debug.log fully clean (zero error lines) across every test session after all fixes above.

Also spot-checked: help (usage message with no args, as expected), who, uptime, version, inventory, say. All correct.

One additional finding, likely intentional design rather than a bug (no error signature; see AGENTS.md's "no error signature = design" standard): emote <text> (free-form emote) and mail require a_permissions >= 500 / >= 100 respectively, but NO code path ever initializes a new player's a_permissions attribute at all -- it silently defaults to 0. Every OTHER bin/user/* command (look, go, say, inventory, drop, get, tell, etc. -- 30 of 33 total) requires exactly permission 0 and works fine for a brand-new player; only these two are gated. Given the MOTD's own text ("added a security system (admin/promote, position, and so commands)") and that emote is the one command letting a player type completely unmoderated free text (unlike the canned-message temote/semote, which both require only 0), this reads as a deliberate 1990s-era anti-abuse gate requiring an admin to promote a vetted player before they get free-text emote/mail access -- not a registration bug. Left unmodified; documented here per the review standard rather than "fixed."

6. WASM

Boots and plays identically under the WASM build (~/src/fluffos/build-wasm/src, via scripts/wasm_client.js): full password-login (reusing the same seeded account) -> look -> quit, byte-for-byte the same output as native (modulo the expected WASM startup noise/mojibake prefix before the listener is ready). wasm_status: playable.

7. Admin account seeding detail

See README.md's "Admin account" section for the procedure. Key files touched: adm/etc/groups (added fluffos to the (admin) group, alongside the original truilkan:cynosure:sulam -- not replacing them), data/users/f/fluffos.o (the per-account "pith file" PLAYER_D maintains independently of the main save; its a_filename attribute, key 300, decides which player-body class std/user/std/maker/std/admin a RETURNING login gets -- critically, ONLY on return-login via check_password(), not on the very first login, since a brand-new account has no pith file yet and always gets "user"), and data/obj/i/admin/fluffos.o (the actual admin-class player save, seeded with a_permissions=2999 (ADMIN(999)) and a_position="Admin", matching the archive's own pre-shipped truilkan admin account's exact values byte-for-byte). Verified: update /bin/user/senses/look succeeds (file write ACL via the groups/access system), and emote (permission-gated, see §5 above) also works once promoted, confirming both independent permission axes (file ACL vs. per-command a_permissions gate) are correctly seeded.

Note: the archive ships a real 1992-era admin account (truilkan, data/users/t/truilkan.o, crypted password unrecoverable) -- left untouched, not reused, per this project's standard policy of never guessing or discarding a pre-existing real account.

8. New AGENTS.md catalog entries added this session

1. §6.2 addition: (: obj_or_path_expr, "method" :) closures only build a real call_other pointer on this driver when the head is a bare compile-time NAME; anything else (string constant, called function) silently becomes a no-op comma-expression. Fix: route through (: call_other, target, "method" :). 2. §6.2 addition: new/class as ordinary identifiers are unconditionally reserved keywords on this driver (object instantiation / typed-class syntax), unlike the compile-flag-gated ref (§6.7). Grep \b(new|class)\b in declaration position on any new archive. 3. §6.2 additions: extract(), log_file(), privp(), cat() are not real efuns on this driver (join the existing tail() entry) -- all reimplementable as simul_efuns per old MudOS manual semantics. 4. §7 candidate: a vestigial private inherit of an otherwise correctly-inherited base class can create an ambiguous SECOND copy of a needed public method, with this driver resolving unqualified calls to the wrong (private, uninitialized) copy -- check whether the private inherit is actually used before "fixing" by loosening its visibility. 5. §4.2 item 4 confirmed on new content: a convert_lib.sh .c->.lpc literal-string rename left a fixed-width extension slice ([len-2..len-1]) unwidened, permanently breaking a hand-rolled command-file cache -- every command in the game failed. Worth a proactive grep (\[\(?len\s*-\s*[0-9]+\)?\.\.) on any archive with a similar directory-scanning command cache.

9. §10.7 deep functional test (2026-08-31)

Full continuous-session playthrough on the native driver (port 40261), exercising literally every implemented bin/user/* command (not just the register/look/go/quit path already verified in §5), plus a maker and admin pass logged in as the seeded fluffos account, plus a WASM long-sit boot watch (scripts/wasm_boot_watch.sh basis 200). This is a genuinely thin lib by design (per the archive author's own README), so "deep test" here meant "does every implemented feature actually work," not a full RPG progression. Found and fixed 7 distinct, previously undetected programming bugs -- more than any prior pass on this lib, all invisible to the boot-log/registration-smoke-test bar because they only fire when a specific command is actually typed.

Bugs found and fixed

1. bin/user/senses/look.lpc: look <item> (without the literal at) crashed on literally every attempt, look.lpc:19-20. sscanf(args, "at %s", something) leaves something unassigned (0) whenever args doesn't start with "at " (i.e. any plain look dictionary instead of look at dictionary); the unguarded present(something, this_player()) right after then hard-errors *Bad argument 1 to present(), Expected: string or object Got: 0, swallowed silently for the non-wizard player (§7.171) -- the client saw NOTHING at all, not even a fail message. Fixed by only attempting the present() lookup when sscanf actually matched (== 1 && something), falling through to the existing "Nothing like that around here." message otherwise -- preserves the room-description's own documented UX ('look at dictionary') while eliminating the crash. New AGENTS.md class, §7.181. 2. 13 bin//std/bin/ command files silently locked at MAX_PERMISSION (999) for every player, including ones whose own permissions() correctly declares 0: finger.lpc, alias.lpc, unalias.lpc, tell.lpc (user-level); clone.lpc, update.lpc, ref.lpc (maker-level); new_master.lpc, galias.lpc, gunalias.lpc (admin-level); bug.lpc, idea.lpc, typo.lpc (report-based). Each overrides create() without calling ::create(), so /adm/std/security/bin.lpc's own create() (seteuid(ROOT_UID); set_permission((int)this_object()->permissions());) never runs -- permission stays at its raw declared default (999) instead of the command's own intended value, and the command's euid stays unset (0) instead of ROOT_UID. Symptom: tell, alias, unalias, finger all failed with "bin: Permission denied." for a brand-new player; bug/idea/typo failed the same way AND (even after the permission fix) their reports never got logged because of bug #6 below. Fixed by adding ::create(); as the first statement in each. New AGENTS.md class, §7.182. 3. 20 command files' actual entry point was named something other than do_command, the one name /adm/std/security/bin.lpc's execute() calls -- every one of these was 100% dead for every player, failing with *Undefined function called: do_command (swallowed for non-wizards, so the player just saw the driver's generic "Errmm?" fail message, indistinguishable from a genuinely unrecognized command). converse.lpc (do_converse); maker-level force.lpc (cmd_force), review.lpc (cmd_review), gauge.lpc (cmd_gauge), popd.lpc/pushd.lpc (cmd_popd/cmd_pushd), snoop.lpc (cmd_snoop), tsh.lpc (cmd_tsh), load.lpc (cmd_load), home.lpc (cmd_home), whatis.lpc (cmd_whatis), origin.lpc (cmd_origin), apropos.lpc (cmd_apropos), man.lpc (cmd_man); user-level give.lpc/remove.lpc (cmd_give/cmd_remove), nickname.lpc/unnickname.lpc (cmd_nickname/cmd_unnickname), history.lpc (cmd_history). Confirmed each renamed identifier is used nowhere else in the archive before renaming (safe, no external callers). This is clearly leftover TMI/Portals-lineage naming convention (the archive's own README says "many of the bin commands are based on the ones from the old TMI") that never got reconciled with bin.lpc's actual do_command contract -- worth checking on sight in any TMI-derived archive. New AGENTS.md class, §7.183. 4. bin/user/objects/ready.lpc: bare ready (no argument) crashed the same present()-with-0 way as bug #1, ready.lpc:15 (arg is 0 when the player types no argument at all, not merely sscanf-unassigned). Fixed with a stringp(arg) guard printing "Ready what?\n" first (matching drop.lpc/get.lpc's existing, correct pattern for the same efun). 5. log_file() (the simul_efun reimplementing the classic MudOS log-append efun, adm/obj/simul_efun/log_file.lpc) silently wrote NOTHING, every single call, project-wide -- confirmed via instrumentation: write_file() inside it always returned 0 (denied) with zero visible symptom (no catchable error; classic MudOS write_file()/its simul_efun reimplementation just returns false on denial). Root cause: this driver's valid_write() master apply receives current_object() as its "user" argument (AGENTS.md §7.176) -- since log_file() calls write_file() directly from ITS OWN code, current_object() at that point is the simul_efun object itself (/adm/obj/simul_efun), whose own euid is 0 (never set), not the actually-privileged calling command (confirmed ROOT_UID euid) that previous_object() would have shown. Every caller was silently affected: bug/idea/typo reports (never appended to log/Bugs/Ideas/Typos despite the interactive editor completing normally), tell.lpc's own error log, master.lpc, and 5+ adm/daemon/inet/* daemons' logging. Fixed by adding seteuid(ROOT_UID); inside log_file() itself before the write_file() call, replicating the "trusted native primitive" semantics the real MudOS efun had. New AGENTS.md class, §7.184 (a direct, simul_efun-specific consequence of §7.176). 6. bin/maker/help/apropos.lpc: APROPOS_D pointed at the stale path /bin/daemon/aproposd, while the real file lives at /adm/daemon/aproposd.lpc -- the exact same archive-internal path-drift shape already documented in §3 for the 19-file bin_m//adm/std/security/bin case, just a single fresh instance. Symptom: apropos <anything> crashed the driver's own master/virtuald fallback into the self-referential *Inherit chain too deep: > 30 recursion (same mechanism as §3's closure bug) on literally every invocation, for every maker. Fixed by correcting the macro's path. 7. std/i/shell.lpc never inherited std/i/nicknames.lpc -- unlike the genuinely-incomplete, never-finished std/i/history.lpc and std/i/tsh.lpc siblings already documented in §4 (both reference undefined variables or #include headers that never existed), nicknames.lpc is a complete, internally-consistent, correctly implemented mixin (query_nicknames/set_nicknames/ query_nickname/remove_nickname/empty_nicknames, all self-contained) that was simply never added to the inherit chain. Symptom: nickname <nick> <real> crashed with *Value being indexed is zero (call_other of an undefined function on the player returns 0, then nicknames[nn] indexes that 0), and unnickname <nick> always reported "No such nickname defined." even right after a (crashing) nickname attempt. Fixed by adding inherit "/std/i/nicknames"; to std/i/shell.lpc alongside the existing pager/edit/alias mixins. Left deliberately incomplete: the mixin's own do_nicknames() substitution helper (meant to actually replace a nickname with the real name in a typed command line) is private and still never called from process_input() anywhere -- wiring that in would mean completing a never-finished feature integration rather than fixing an omission, so nickname/unnickname now correctly manage the nickname list but the substitution effect itself remains unverified/unimplemented. Documented here rather than guessed at.

Confirmed NOT bugs -- genuinely incomplete pre-existing content (left alone)

• history command (bin/user/shell/history.lpc, entry point now correctly renamed per fix #3 above) depends entirely on std/i/history.lpc for query_max/query_cmd_num/query_ptr/ query_history -- but that mixin is the SAME file already documented in §4 as "genuinely incomplete original content... referencing ptr/max/cmd_num variables that are never declared anywhere," never inherited by anything live. Now that the entry point is reachable, typing history silently no-ops (the undefined call_others return 0, max ends up 0, the display loop's range is empty) instead of crashing -- a mild improvement, but the actual feature remains unimplemented by the original author, not by this port. Left as-is, matching the existing precedent exactly.
• popd/pushd commands (entry points now correctly renamed per fix #3) depend entirely on std/i/tsh.lpc for their actual popd()/pushd() methods -- but tsh.lpc #includes <commands.h> and <tsh.h>, both of which "never existed anywhere in the archive" (§4's own words, re-confirmed here), so it cannot even compile, let alone be inherited. Typing popd/pushd now correctly reaches do_command() (fix #3) but silently no-ops when previous_object()->popd(str) calls a function that exists nowhere on any real player object. Left as-is, same bucket as history/ tsh above -- this is the SAME already-abandoned shell rewrite ("Brutally hacked up and destroyed by Buddha... on 3-7-92"), not a fixable omission.
• emote <text>/mail permission gate (§5's original finding): re-confirmed still true and still reads as intentional design, not touched.

WASM

scripts/wasm_boot_watch.sh basis 200 ran the full 200s with no driver crash, reaching the login prompt cleanly. One expected, already-documented WASM-only limitation fired exactly as AGENTS.md predicts (line ~349, "WASM builds ship without sockets... package"): adm/daemon/inet/inetd.lpc fails to compile under WASM (Undefined function socket_close/socket_write/socket_create/ socket_error), so /adm/daemon/inet/inetd never loads and the intermud-tell/finger feature is unavailable under WASM specifically -- master.lpc's own socket_preload() already wraps this in catch(), so it's non-fatal and the rest of the game is unaffected. Native-only limitation, not a bug; not fixed.

Regression check

All 7 fixes re-verified together in one final continuous session (register deeptest16, exercise every fixed command, quit) plus a separate maker/admin pass logged in as the seeded fluffos account exercising every one of the 13 previously-dead maker commands (apropos, review, force, gauge, popd, pushd, snoop) -- zero errors in the driver's own captured stdout across the whole session (native builds' debug.log is dead for the process's whole life per §10.9, so stdout is the only reliable error channel).