Two independent reviews of the seeded-create change found three more
places where the newly durable hidden row, or the new create-time copy,
was not handled by the same rule as the rest of the path:
- Message search (dashboard search and the session_search tool) had no
display_kind filter, so a hidden opening row matched a query the
person never saw. The shared search predicate now skips hidden rows.
- _seed_row left the fresh session row behind when the transcript copy
failed after the row was committed. The first prompt's retry copies
the whole seed, so a kept partial copy would be duplicated. The row
is now deleted when the copy did not complete, the compensation
_persist_branch applies to branch children; the first prompt then
starts clean.
- _live_session_payload (a resume that reuses a live session) reported
message_count as the raw history length while its messages array was
filtered. It now follows _resume_response: the stored size when
messages are omitted, else the wire count.
Tests: the two seeded-create tests now drive the first-submit path
through _persist_session_row_for_submit, the function prompt.submit
calls, and assert search and the reuse-live count; a third test pins
the rollback (no row after a failed copy, one copy after the retry).
* fix(tui-gateway): a seeded session is durable at create, and its seed is written once
session.create accepts opening messages. Three defects sat in that path:
- A seeded session without a parent was never persisted at create, so a
restart before the first prompt lost it and session.resume answered
4007. Only branch children (#93959) were persisted up front. The
same rationale applies to any seeded create: seeded content is
intent, not an abandoned draft. Parentless seeds now persist their
row, transcript and client title at create; empty drafts stay lazy.
- _coerce_seed_history dropped display_kind, so a seeded row tagged
"hidden" (model-facing scaffolding) rendered as a user bubble. The
coercion keeps "hidden" and only "hidden"; every other kind is
stamped by the gateway at turn time and is not accepted from the wire.
- A branch child's seed was written twice: _seed_branch_row copied it at
create but never marked it persisted, so the first prompt's
_persist_branch_seed appended the copy again. The create path now
sets _branch_seed_persisted, and the gate is a create-time `seeded`
stamp instead of parent_session_id, so a resumed session (whose
history comes from the DB) can never re-append its transcript.
Two invariant tests, both red on main: a parentless seed survives a
gateway restart with the hidden row kept out of the wire transcript and
not re-written by the first-submit path; a branch child's seed is stored
exactly once. The reasoning-fields fixture stamps `seeded`, the flag
session.create sets.
* fix(tui-gateway): a hidden seed row stays out of the list preview and the create count
Live-testing the seeded create on every surface showed two places where
the newly durable hidden row (display_kind="hidden") still surfaced:
- session.list built a session's preview from its first user row with no
display_kind filter, so a hidden opening row (model-facing scaffolding
the gateway never paints) became the sidebar preview. The preview
predicate now skips hidden rows, in every listing query that shares it.
- session.create reported message_count as the raw seed length while its
messages array already filtered the hidden row (2 vs 1). It now counts
what is on the wire, the same rule session.resume applies.
Both are covered by the existing seeded-create test: the create count
equals the wire transcript, and the preview of a session whose first
user row is hidden is its first visible user row.
* fix(tui-gateway): a live unpersisted resume counts the wire transcript
session.resume on a live session that has no row yet reported message_count as
the raw history length while its messages array was already filtered, the same
mismatch the previous commit fixed on session.create. Count the wire, as the
cold, deferred and reuse-live resume paths already do.
* chore: retrigger CI (zero-job dispatch failure, auto-heal)
Keep the isolation contracts (bot never on the human peer, bot turn never in the
human session, writes refused mid bot-turn, one join per author, signature busts
the cache) and drop the alias/prefix/sanitize enumerations. 795 -> 454 lines.
sync_turn reads a2a_sessions from the config bound when the provider was built. A cached gateway provider kept the old value after honcho.json flipped it, because the signature that busts that cache did not carry the flag.
_joined_author_peers kept an entry for every honcho session the manager ever wrote to. It now holds at most _SESSION_CACHE_MAX_SIZE sessions and drops the oldest past that, so a forgotten session's authors rejoin on their next write. A failed join no longer leaves an empty entry behind.
The join read the manager-wide user_observe_me and user_observe_others directly. It now asks _join_observation_flags(honcho_session_id), which returns the same values today. #103889 stores the effective flags per session and replaces the body of that method.
_generated_runtime_peer_id takes a reserved set, and the bot path passes the session's human peer ids: each runtime id and the peer _resolve_user_peer_id returns for the key. bot:coder with a runtime human coder and no runtimePeerPrefix now gets the digest suffix. _explicit_user_peer_ids keeps its meaning for prefixed runtime users.
Two senders named coder on different connections get different peers and different a2a sessions, and a userPeerAliases entry keyed by the full connection-qualified id wins.
identity_signature now carries cfg.workspace_id. The gateway folds these values into its agent cache key, and a workspace change in honcho.json reused a cached agent that was still bound to the old workspace.
_a2a_session_key now names the session <session>:a2a:<aiPeer>:<sender id>-<digest>. Two profiles that share a workspace and a session key wrote one sender's DMs into one Honcho session. The recipient peer comes from the same aiPeer derivation the session builder uses, moved into session_peers.assistant_peer_id_for so the two cannot drift.
_peer_id_for_runtime_id now looks up userPeerAliases by the full bot id and otherwise passes everything after bot: through _generated_runtime_peer_id. A digest suffix is added when sanitizing changed the id or the result equals peerName or an alias target, so bot:eri never resolves to the operator's peer and bot:a.b stays apart from bot:a-b. The docstring and README no longer claim a cloned profile's aiPeer defaults to the profile name.
The plugin spelled the `a2a:` prefix itself. `agent.turn_author.a2a_key` is the shared name
for a bot author's turns, so the session key now derives from it and every reader that files
bot turns apart agrees on the prefix. The resulting key is unchanged.
A gateway platform marks a bot sender with its raw user id and a bot flag, never a `bot:` id.
`resolve_author_peer_id` treated that author as a human, so `pinUserPeer` collapsed a bot onto
`peerName` and an unresolved peer opened the a2a session under the human's peer. The resolver now
takes `is_bot` and gives every bot its own peer. `sync_turn` skips the turn when no peer resolves
or when the peer equals this agent's `aiPeer`.
The a2a session key carries an eight-character digest of the author id, so two ids that sanitize
alike stay in separate sessions. During a bot-authored turn `honcho_conclude` and `honcho_profile`
refuse writes and the built-in memory mirror is skipped, because conclusions and cards describe
the human. The README paragraph on bot DMs now matches the code.
with pinUserPeer on, resolve_author_peer_id returned None for every author,
so a bot dm's words were written under the human's pinned peer inside the
a2a session. the pin exists to unify one person's platform accounts. a bot
is not one of them.
bot: authors now resolve to their peer before the pin check, so a pinned
operator still gets bot speech attributed to the bot.
A DM relayed from another Hermes profile ran as a turn in the recipient's
Bot Chat session. sync_turn wrote the bot's words and the recipient's reply
into that session, and before per-author writes they landed under the
human's peer. The human's representation absorbed conversations the human
never had.
The turn context now marks such turns with scope a2a:<bot id>.
sync_turn routes a bot-authored turn into a separate Honcho session keyed
<session>:a2a:<sanitized bot id>, created with the sender bot as its user
peer, and never writes it into the human's session. The key is deterministic
so every turn from the same bot reaches the same session, and it stays
inside Honcho's 100 character session id limit. Recall still reads the
human's session only.
a2aSessions (host block, then root, default true) turns the routing on.
With it off, bot-authored turns are skipped. A bot turn that names no
author id is skipped as well, because nothing can key its session. Human
turns are unchanged.
get_or_create takes a user_peer_id override so the a2a session's roster is
the bot and the assistant, not the runtime human.
The gateway agent cache read honcho.json itself through a honcho-named
block in gateway/run.py and gateway/run_agent_cache.py. Every other memory
provider had no way to bust the cache when its identity mapping changed.
HonchoMemoryProvider.identity_signature() now returns the same values under
provider-neutral keys: user_identity, agent_identity, pin_user_identity,
runtime_identity_prefix, user_identity_aliases, session_prefixing. The
gateway files them under memory.<key> through the MemoryProvider hook. The
hook reads config only, memoizes on the file's mtime and size, and returns
an empty dict when the file cannot be read.
The honcho-specific extractor, its memo and its key tuple are gone from the
gateway. The pinPeerName cache-busting test now asserts on
memory.pin_user_identity.
The bot-mode dispatcher names another profile as bot:<profile>. The
resolver treated that like a human runtime id, so a configured
runtimePeerPrefix produced peers like telegram_bot:coder and a profile
that already owns an AI peer in the same workspace got a second one.
A bot:<profile> author now resolves in this order: a userPeerAliases entry
for the full bot id, else the sanitized profile name. A cloned profile's
aiPeer defaults to the profile name, so a same-workspace sender lands on
its existing AI peer. Prefixes never apply to bot ids. pinUserPeer still
collapses bot authors onto the pinned peer, the same as every other author.
sync_turn only knew the author through the on_turn_start stash. The memory
manager now passes turn_author and scope with the turn, and a caller
that skips on_turn_start left the stash empty or stale.
sync_turn takes both keywords and reads the author from turn_author first.
The stash stays as the fallback for callers that never pass it.
resolve_author_peer_id compared the author against the primary runtime id
only. A transport that names the participant by the alt id (Telegram
username instead of UID) got a second peer for the same person. Either
runtime id now counts as the session's own participant.
The manager resolved one user peer in `get_or_create` and froze it onto the
session, then `_flush_session` chose between it and the assistant peer by
role. Every user turn in a shared session landed on that one peer, so the
first person to message the agent collected everyone else's facts — and a
Honcho conclusion, once derived, is not self-correcting.
`resolve_author_peer_id` maps the turn's author onto its own peer using the
alias-then-prefix order `_resolve_user_peer_id` already applies, so an
aliased account reaches the same peer whichever turn it wrote. `sync_turn`
resolves it before starting the write thread, so a following turn cannot
retag a queued write. `_flush_session` then writes each user message under
that peer.
A shared session's roster is open — people and other agents arrive after
the session exists — so `_author_peer_for_session` joins a peer when it
first writes instead of enumerating participants at init. Joins are
remembered per session, and a failed join still writes under the right
peer, losing only the observe config.
Three cases return None and keep the session's own peer: no author named,
the author IS the session's peer, and `pinPeerName` set — that flag is an
explicit request to unify identities, so it still collapses authors in a
shared chat.
An unnamed author stays unattributed rather than defaulting to the owner.
That preserves today's behavior for the transports that send no author, so
those turns still reach the session peer; #83500 owner-gated the memory-file
migration for the same reason.
Display names never become peer IDs — they are attacker-influenceable on
any platform where participants set their own name.
Found by using the feature as a user (natural prompts, real sites, CLI PTY + native Electron), not by naming tools:
- browser_vault_save_login was registered but never offered: toolsets.py is a hand-maintained list. Added, with an
invariant test that every registered browser_vault_* tool is in the browser toolset.
- Vault tools were absent on the DEFAULT backend (Browser Use): the gate deferred to check_browser_requirements(),
which is False by design there. Gate = is_browser_use_cli_mode() or check_browser_requirements().
- The model typed a page-shown demo password with browser_type and offered to take one in chat: the vault rules
lived only on the vault tools. browser_type/browser_exec now carry a vault note when the vault tools are
present ("call browser_vault_list first … never type a password with this tool, never accept one in chat, even
if the page shows it"); the browser_exec login-wall line points at the vault instead of "ask the user".
- On Browser Use the saved item was bound to chrome://new-tab-page: the supervisor's default page session is the
daemon's blank tab. browser_vault_save_login now focuses the tab holding a password field before reading its
origin (focus_page("", accept=probe); about:/chrome: pages are never candidates). Live E2E leg added.
- Settings row: "identifier · Added <date>", origin omitted when it duplicates the label.
Live (real model): CLI on Browser Use — first visit prompts, signs in, saves; second visit fills silently; GitHub
decline (Enter or ESC) stops the agent, which refuses chat passwords. CLI on the built-in stack — same three
scenarios pass. Desktop native Electron — same three scenarios plus Settings list/remove pass. Password never in
a transcript, UI, or a file outside vault/.
Nobody should have to learn `hermes vault add` or find a toggle before "log into GitHub" works.
- browser_vault_save_login: when the agent reaches a sign-in page with no saved login it asks the user
on THEIR surface (CLI two-step panel on the sudo modal: identifier shown, password masked; Desktop
card with labelled Email/username + Password fields). The answer goes to the encrypted vault bound to
the page origin and is filled at once; the model gets back only the handle and identifier. Declining
returns save_declined; headless sessions get prompt_unavailable. Never a password in chat.
- Vault tools ride with the browser toolset (check_browser_requirements) instead of appearing only once
the vault has items — an empty vault is exactly when save_login is needed. browser_vault_list hints
at it when empty.
- 1Password / Bitwarden are login sources as soon as their CLI is installed; `vault.<name>.enabled`
is opt-OUT only. Settings shows Detected/Locked/Unlocked/Off/Not detected with a switch only for
installed managers; `hermes vault sources` reports detection, `--disable`/`--enable` flip the opt-out.
- Desktop nav/page renamed "Passwords & Logins"; empty state tells the user they do not need to add
anything; all five locales updated. Docs rewritten from "how it works" to "say log into X".
- New per-thread SaveLoginPrompt callback (agent/vault_backends/unlock.py) installed beside the unlock
prompt on every CLI site and the gateway bridge (vault.save_login.request/respond/expire), propagated
to worker threads via tools.thread_context.
Live: CLI PTY (real model, packaged Chromium, local login server) — panel shown, identifier + masked
password typed, server received the correct password, password absent from terminal transcript and
from every file under HERMES_HOME outside vault/. Native Electron (headless, isolated HOME/HERMES_HOME,
own Vite + CDP port) — card shown, "Save & sign in", server received the password, Settings lists the
saved item, password absent from the rendered UI.
Found by a model-driven live run (hermes chat -q against a real login page): the agent found the
vault tools, typed the identifier, then failed twice for reasons the direct-call E2E could not see.
- The three schemas spelled their arguments `input_schema` (Anthropic shape). The registry and every
provider adapter read `parameters`, so the model was shown browser_vault_fill with NO arguments and
called it with an empty handle. Renamed; a registry-wide invariant test now fails on any schema
without `parameters`.
- A local built-in session (agent-browser --session) carries no cdp_url, so nothing ever started a
supervisor for it and the fill refused with supervisor_required. _ensure_supervisor asks the daemon
for the packaged Chromium's endpoint (`get cdp-url`, carries no secret) and attaches on demand; the
fail-closed test now pins that the daemon is only ever asked `get`, never handed an eval with the
password.
Live: same run after the fix -> browser_navigate, browser_vault_list, browser_type, browser_vault_fill,
browser_click; the test server received the correct password; landing title "Welcome"; the password
string is absent from the whole transcript.
payment and address items could be stored (CLI wizard, Desktop dialog) but nothing could
fill them: a dead surface holding real card numbers. browser_vault_fill now handles all
three kinds through the same origin-bound, supervisor-only, redacted path:
- classify_checkout_control / select_checkout_fills map WHATWG autocomplete tokens
(cc-number, cc-exp[-month|-year], cc-csc, address-line1/2, address-level1/2, postal-code,
country-name) with label/name heuristics as backup; a combined "MM/YY" control gets
exp_month+exp_year and suppresses the split fills; inspection now covers <select>
(country, state, expiry month) and the fill script picks an option by value or text.
- Every payment fill goes through request_elicitation_consent (gateway button round-trip
or CLI panel) before a byte is written; declined → payment_declined, headless sessions
are refused. A prompt injection that reaches a checkout can ask, not spend. Card values
join the redaction registry like passwords; the result lists targeted field tokens only.
- Origin is now required for every kind (CLI wizard asks; Desktop dialog always shows the
field) because a card without a bound origin is unfillable.
- The tool descriptions, docs and CLI copy drop "Phase 1 / login only".
Live (evals/vault_fill_live_e2e.py, real browser_exec + packaged Chromium): decline writes
nothing; accept fills card/expiry/CVC on the /checkout tab, leaves the email box and the
country <select> untouched, and neither the card number nor the CVC appears in any result.
browser_vault_tool also: focuses the tab on the bound origin holding the right form before the
origin pre-check (focus_page from the previous commit); tool descriptions say "the browser's
input tool" (rewritten per session by model_tools); _check_vault_available is registered
uncached because its answer is per profile (vault dir + config) and the probe is a file stat.
On the default backend (browser.backend unset → browser_exec) the vault tools were
advertised but could never fill: the CDP supervisor that carries the secret-bearing
eval is started only by the built-in browser_* session path, so _eval_js_secret
failed closed with supervisor_required and the origin pre-check fell back to an
agent-browser CLI eval against a browser browser_exec never touched.
- browser_exec now attaches SUPERVISOR_REGISTRY to the CDP endpoint it just routed
the harness to (BU_CDP_WS/BU_CDP_URL), so the fill talks to the SAME browser over
the same secret-capable WebSocket. BU direct-cloud (BU_AUTOSPAWN) exposes no
endpoint and keeps the supervisor_required refusal.
- CDPSupervisor.focus_page(origin, accept=<js>) (used by browser_vault_fill, next commits) re-attaches the page session to the
open tab on the item's origin whose DOM holds the form being filled (browser_exec
opens its own tabs; the supervisor's initial attach picks the first page target,
which is chrome://new-tab-page). browser_vault_fill uses it before the origin
pre-check with a per-kind probe (password input / card fields / address fields).
Live: evals/vault_fill_live_e2e.py drives the real browser_exec tool against Hermes'
packaged Chromium with the login page in the third tab; A/B with the attach line
disabled fails at "did not attach a supervisor", enabled fills the password into the
/login tab and card fields into the /checkout tab with every model-facing read
scrubbed.
Also: browser_vault_list/fill described the workflow as "type the identifier with
fill_input", a helper that exists only inside browser_exec code (toolset browser-use) and is
a ghost on the built-in stack. model_tools._rewrite_browser_vault substitutes the concrete
name from the session's actual tool set (`fill_input` inside browser_exec, or browser_type),
the same dynamic cross-reference pattern browser_navigate uses for web_search.
browser_navigate / browser_exec / web_extract refused any URL whose query carried a
credential-NAMED parameter (token, signature, access_token, ...) when the backend was
a cloud provider. That is exactly the shape of magic links, OAuth callbacks and signed
CDN assets, so on Browserbase/Browser Use the agent could not finish a sign-in flow or
open an X video asset ("Blocked: URL contains a credential-like query parameter").
The floor protected nothing: the cloud browser already sees every cookie and typed
password of the session, and with the credential vault it receives the real password at
fill time. Hermes' own secrets leaking into a URL stay blocked by the value-shaped
_PREFIX_RE check (_secret_url_error), which is backend-independent. IMDS and
private-address floors are unchanged.
Manager tokens: lock generation fence (a Lock acknowledged while `bw unlock`
/ `op signin` is still running discards the late token); tokens record the
unlocking gateway session and are released when THAT session ends, not when
any sibling session in the profile is torn down.
1Password: OP_CONNECT_HOST/TOKEN come from the profile's scoped secret store
like the service token (Connect outranks a service token inside op), never
from the launch environment.
Vault RPCs bind params.profile (home + secret scope) so a shared remote
backend serving several profiles locks/lists/unlocks the requested one;
unknown profile → RPC error, not a crash.
Fill target: inspection stamps are `<nonce>:<index>`; a fill resolves only
its own inspection's stamps, so an interleaved second inspection can no
longer redirect A's password into a newly mounted field (real Chrome: 0
filled, both fields empty).
Desktop Settings: every RPC goes through the owner profile's socket
(requestGatewayForProfile), query keys carry (connection, profile), an owner
change closes dialogs and wipes drafts (a master password typed for A is
never submitted to B; a late list from A never paints under B), and vault.add
secrets travel in a ref consumed by the mutationFn instead of mutation
variables. Three owner-routing invariant tests on the real component.
Docs/PR body: session-scoped release, lock-race semantics, bw --passwordenv.
Bitwarden unlock now uses the CLI's documented non-interactive channel:
`bw unlock --raw --nointeraction --passwordenv VAR`, VAR set on the child
environment only (bw 2026.x rejects a piped password with "Master password
is required"). Verified against the real published binary.
Manager session tokens are keyed by (profile home, backend): a Desktop
gateway hosting several profiles can no longer reuse or lock another
profile's session. Status probes (`vault.sources`, is_unlocked) no longer
refresh the idle TTL; only real manager calls do. Gateway session teardown
locks the profile's managers (a per-session unlock ends with the session).
1Password service-account token comes from the profile-scoped secret store
(get_secret), not ambient os.environ.
`vault.source.set` no longer references a module constant (bind_module
rebinding dropped it → NameError on every Settings toggle).
Fill target binding: inspection stamps each input with a per-inspection
slot attribute; the fill resolves by stamp and requires type=password, then
strips every stamp. A DOM reflow between inspect and fill can no longer
redirect the password into a text field (reproduced in real Chrome before,
0 filled after).
Redaction boundary: no 4-char floor, CR/LF-normalized form registered
(what a text input actually stores), JSON object KEYS scrubbed in both
browser redactors; longest value first. Docs now state the real trust
model: accidental-disclosure protection, not an execution sandbox.
Desktop: the mid-turn card sends the master password through the owning
session's socket (requestForOwnedSession), never the ambient foreground
gateway; `vault.unlock.expire` clears a stale card; Settings keeps the
master password out of react-query mutation variables (ref consumed by the
mutationFn). One renderer invariant test for the routing.
_callback_api() now yields (getter, setter) pairs for every per-thread prompt
(approval, sudo, vault unlock); the kernel cell captured and restored the old
fixed 4-tuple. Iterate the table so a cell carries every callback and a future
addition needs no change here. Test recorder unpacks the new shape.
Also: perfectionist import order in ui-tui interfaces.ts (CI lint).
The browser vault now draws from three login sources behind one handle
shape: the local encrypted vault (vault_…), 1Password Login items (op:…)
and Bitwarden Password Manager logins (bw:…). browser_vault_list aggregates
metadata across them; browser_vault_fill routes by prefix and resolves the
password at fill time only, through the manager CLI.
External managers are locked until the user unlocks them for the current
session. The new browser_vault_unlock tool (and the fill path, implicitly)
asks the surface to show a masked master-password prompt — CLI panel
(reuses the sudo panel state), TUI/Desktop via a vault.unlock.request
blocking card. The password goes to `op signin --raw` / `bw unlock --raw`
on stdin, never argv or env; only the session token is kept, in memory,
with a 30-minute idle TTL, cleared on session close or `vault.lock`.
Headless contexts (cron, webhook, api_server, -q) can never prompt: the
manager is reported as locked with unlock=unavailable_in_this_session and
fill refuses — the same posture approvals take where nobody can answer.
Config: vault.onepassword / vault.bitwarden {enabled, binary_path, …};
a 1Password service-account token skips the prompt for headless use.
RPC: vault.sources, vault.source.set, vault.unlock, vault.lock for Settings.
Tests (2, real subprocess against a fake bw; each proven red by sabotage):
headless never prompts or spawns; unlock feeds stdin only, token never
enters os.environ, fill routes by prefix and the password only reaches the
fill script.
Consolidated re-apply of #96988 onto current main. Ported from
Merit-Systems/OpenInstinct (MIT) opaque-handle autofill design: the model
sees vault handles + login metadata, the password is resolved and filled
server-side over the supervised CDP socket, and filled values are scrubbed
from every browser tool result by an unconditional redaction registry.
Rebase adaptations to the Sep-2026 facade/sibling layout:
- toolsets: one _HERMES_CORE_TOOLS entry (the browser toolset derives from it)
- hermes_cli/main.py: vault parser registered via the subcommand owner table
- file_safety: vault/ joins the _READ_DENIED_DIRS credential-dir table
- redact: registry scrub runs before the redact_secrets early-return
- browser_vault_tool: _run_browser_command now lives in browser_tool_session
The facade forwarded turn_author= to the loop's public entry point, which did not
declare it: every real AIAgent.run_conversation() turn raised TypeError (16 CI
failures across provider, sidecar, cron and finite-chat suites). The PR's tests
only exercised build_turn_context directly, so the missing hop was invisible.
Adds one facade-through-loop test that goes red when the kwarg is dropped.
Keep one or two behaviour tests per seam (author reset on a cached agent, forged
_turn_author refused, guard trips and cools, single charge on the busy path) and
drop the parser/setting enumerations. a2a_key goes with them: nothing in this PR
reads it; the honcho follow-up that does can bring it back with its consumer.
bot_relay.deliver accepted from_profile, from_handle and from_connection from any admitted JSON-RPC client. The handler now refuses them with error 4095 when the calling transport carries a browser login identity, since a logged-in browser never relays for another connection. The DeliveryAuthor docstring now says the author is trusted because an admitted client relays it, not because the sender is verified.
message_agent sent a bare bot:<profile> id through hermes peer dm, so a remote coder and the recipient's own coder shared one author id. The peer branch now sends bot:<hostname>/<profile>, with the hostname cleaned like any author field and slashes dropped. The direct local path keeps the bare id.
The author travels on the queued entry and the _run_prompt_submit call, and turn_context resets agent._turn_author at every turn start. The test pins that the human prompt following a drained relayed dm reaches run_conversation without a turn_author.
delivery_turn_author kept the bare bot:<profile> id when the sender's connection was the Desktop's own "local", so a DM relayed from that machine collided with the recipient's profile of the same name. A relayed DM always crosses gateways, so the connection id is now part of the id whenever the Desktop sends one, and only the direct message_agent path in tools/bot_mode_dm.py stays bare. The relay.ts and session_auto_continue.py comments added earlier are cut to one line each.
A relayed DM stamped bot:<profile> on the recipient turn, so an ops profile on another machine and the local ops profile shared one author id. The Desktop now forwards from_connection with each bot_relay.deliver, and delivery_turn_author builds bot:<connection>/<profile> for it while the Desktop's own gateway ("local") keeps the bare id. An api author object accepts an optional origin string that yields the same shape.
MemoryManager.on_turn_start forwarded the author kwargs to every provider and _each_provider swallowed the TypeError, so a provider with the two-positional on_turn_start(n, text) stopped running. The kwargs are now filtered against the provider's signature the way _provider_sync_accepts filters sync_turn.
_enqueue_prompt dropped a queued entry whose text matched the live prompt without checking turn_author, and _sanitize_queued_entry_vs_inflight_user stripped the inflight prefix from any text entry. A relayed DM carries its sender, so an entry with turn_author is never treated as a self-duplicate and never prefix-stripped.
The loop guard peeked under the transport profile stamped on the source but counted under the routed profile's scope, so the two read different gateway.bot_loop_guard blocks. _admit_bot_message_for_source runs the count under the same _authorization_profile_home as _is_user_authorized_for_source, and both the ingress gate and the busy path call it.
_handle_active_session_busy_message only peeked at the guard, so a bot steering a running turn was never counted and the budget never tripped. The busy handler now admits the message once after the authorization verdict and marks the event. _hm_admit_event skips a marked event, so a follow-up that was queued on the busy path and drains later is not charged twice.
When the target Bot Chat is already open on this gateway, the relay handler delivers through
`prompt.submit` with `queued: true`, and that branch dropped the envelope's sender. The model
still saw the text prefix, but the turn reached the agent unattributed, the exact case the
subprocess branch fixes.
The relay handler now stamps the author on the submit as a `DeliveryAuthor`, an in-process object
a JSON client cannot build, so `prompt.submit` accepts it the way it accepts a hosted-room callback
and refuses a dict with error 4124. The busy queue keeps an authored envelope in its own slot, the
drain hands the author to the turn runner, and the runner passes it to an agent that declares the
keyword. A plain prompt after an authored dm carries no author.
Local deliveries to a desktop-owned Bot Chat take the live-owner mailbox instead. The admission
intent and the mailbox record now carry the author, a retry under the same id with a different
author is refused, and the owner gateway hands the author to the turn it runs. Isolated compute
turns still run unattributed, because the compute-host frame has no author field.
The gateway turn runner and the quiet one-shot passed `turn_author=` unconditionally. Every
test double and wrapper with an older `run_conversation` signature raised TypeError, which CI
caught across twenty gateway tests. Both call sites now check the callee with the existing
`_accepts_keyword` helper. A human `-Q` turn keeps today's call shape with no author keyword.
The Telegram adapter asks the authorization check before dispatch, the ingress gate asks it
again, and the busy path asks a third time. Each call counted one loop-guard event, so a
Telegram bot tripped the budget after a third of the configured messages. The verdict now only
refuses a chat that is cooling down. The ingress gate counts an admitted bot message once.
`parse_turn_author` treats only booleans, integers and the strings true/1/yes as a bot flag,
and returns None for an author with neither id nor name. Names keep format characters and
non-breaking spaces so emoji sequences survive. The quiet one-shot pops HERMES_TURN_AUTHOR
before the turn so tool subprocesses do not inherit it. `max_events` must be a whole positive
number. Issue numbers move out of code comments.
#91483 added the guard inside the ALLOW_BOTS block. on today's authz layout
an admitted bot returns from an earlier rule before that block runs, so the
guard never tripped (its own ping-pong test passes turn 21 on this tree).
_is_user_authorized now computes the allowlist verdict first and runs every
admitted bot-authored message through the guard, so bots admitted by a chat
allowlist, allow-all, pairing or delegation are metered too. the budget stays
per conversation. settings move from HERMES_BOT_LOOP_* environment variables
to config.yaml gateway.bot_loop_guard (defaults added to config_defaults),
re-read on every call. an unauthorized bot dm no longer gets a pairing code,
so a cooldown produces no outbound traffic. tests rebuilt against real
SessionSource objects.
Telegram Bot API 10.0 (2026-05-08) lets bots receive messages from other
bots, and the platform ships no loop guard of its own. core.telegram.org
/api/bots/bot-to-bot ("Loop prevention") requires the BOT to make
bot-message handling terminate predictably via dedupe, per-chat rate
limits and maximum interaction depth, and warns that "failure to handle
loops properly may lead to degraded performance or platform
restrictions".
Hermes had no brake at all: TELEGRAM_ALLOW_BOTS in gateway/authz_mixin.py
was a bare `return True` with no counter, window or cooldown.
TELEGRAM_ALLOW_BOTS=mentions does not help, because when bot A replies to
bot B the reply itself satisfies the mention test, so every turn re-arms
the peer. Observed 2026-08-21: two Hermes bots exchanged 132 messages in
one group before a human intervened.
This adds gateway/bot_loop_guard.py, a thread-safe sliding-window budget
with cooldown and sweeping of stale buckets, wired into
_is_user_authorized. Defaults (20 events / 60 s window / 60 s cooldown)
match the OpenClaw reference implementation.
Two placement details that are easy to get wrong:
- The guard runs BEFORE the TELEGRAM_GROUP_ALLOWED_CHATS shortcut. That
env var returns True for any sender in an allowlisted chat, bots
included, ~32 lines before the is_bot block, so a guard placed in the
is_bot block never executes for the one configuration that needs it.
- The budget is keyed per CONVERSATION, not per (sender, receiver). This
process only sees inbound messages, so the receiver is constant; keying
on the pair would give each sender its own budget and N bots would need
N x budget messages to trip a guard meant to cap the whole exchange.
Installations without ALLOW_BOTS are byte-identical in behaviour: the
guard only runs on the source.is_bot branch with ALLOW_BOTS set to
mentions or all. HERMES_BOT_LOOP_PROTECTION=off is a full kill switch.
Adds tests/gateway/test_bot_loop_guard.py (7 cases), each verified
failing against the tree without this patch.
POST /api/sessions/{id}/chat, /chat/stream and /v1/runs take an optional
"author" object. absent or null keeps today's behavior; a non-object is a 400
invalid_author. the value reaches run_conversation as turn_author and nothing
else: it is a claim by an API-key holder and grants no capability.
hermes peer dm and peer run add the author to the request body when the
message_agent runner launched them with HERMES_TURN_AUTHOR set, so a
cross-machine dm is attributed the same way a local one is.
a bot dm arrived as an ordinary user message. the only trace of the sender
was the "Message from" text prefix, which the model reads and nothing else
does. the recipient's memory provider saw its own configured user.
message_agent now passes the sender as {"id": "bot:<profile>", "name":
<handle>, "is_bot": true} to the delivery runner (--author <json>), which sets
HERMES_TURN_AUTHOR on the recipient one-shot only. the -Q turn reads it and
passes turn_author into run_conversation. the desktop relay forwards the
envelope's from_profile/from_handle to bot_relay.deliver, which sets the same
variable on its delivery turn. the runner drops any inherited author first so
a delivery without one stays unattributed. the text prefix is unchanged.