Commit Graph

25224 Commits

Author SHA1 Message Date
Brooklyn Nicholson 90ee4460cb fix(desktop): translate sidebar recents_profile through SSH aliases
Managed SSH maps a Desktop profile label onto a different remote name.
The sidebar filter lives in recents_profile, so rewriting only ?profile=
left those reads on the remote default and the Sessions list came back empty.

Co-authored-by: noah <loahnisk@gmail.com>
2026-08-25 12:07:00 -05:00
ethernet cbd8de8ad6 fix(gateway): bind session cwd for the live system-prompt rebuild
The function _persist_live_session_system_prompt runs on the RPC
dispatcher thread. A model switch calls it. On that thread the
_SESSION_CWD contextvar is not set. Because of this,
resolve_agent_cwd() falls back to the process TERMINAL_CWD value.
The desktop pins TERMINAL_CWD to the home directory. The rebuilt
prompt then records the home directory as the working directory.

The wrong prompt persists to the session database. Later turns
restore the stored bytes without change, because the turn prologue
rebuilds the prompt only when the cache is empty. The wrong line
never heals. The terminal tool is not affected, because it reads
the per-session cwd record.

The desktop composer sends a model switch before the first turn
when its model differs from the config default. Because of this,
a new project session can show the wrong working directory for
its full life.

Fix: bind the session context around the rebuild, in the same way
the function already binds the profile home for issue #50233. The
test drives the real function from a bare thread and asserts the
persisted prompt carries the session cwd.
2026-08-25 09:47:36 -07:00
Teknium 1bbb6e5bce docs(skill): troubleshoot stale web_extract pages — cache carveouts + cache_exempt_hosts
Adds a 'web_extract shows a stale page' section to the bundled
hermes-agent skill's troubleshooting reference: the 20-minute TTL
symptom, the automatic never-cache carveouts (localhost/private/LAN,
website_blocklist, rescue/failed responses), the web.cache_exempt_hosts
recipe for staging/tunnel sites, and the TTL/disable fallbacks.
2026-08-25 04:21:45 -07:00
Teknium ba9fc55e16 feat(web): cache_exempt_hosts — always-live fetches for staging/tunnel sites
Sites under active development but tested over the public internet
(Vercel previews, ngrok tunnels, staging domains) are public DNS, so
the local-dev never-cache rule can't catch them. web.cache_exempt_hosts
lists hosts whose pages are always fetched live: exact, "*.wildcard",
or domain-suffix matching (label-boundary aware — mysite.dev covers
preview.mysite.dev but never evilmysite.dev). Checked on both store
and lookup, so adding an exemption takes effect immediately even for
entries cached before the config change.
2026-08-25 04:21:45 -07:00
Teknium f0381ee4ae fix(web): never cache local development URLs in the extract cache
Dev servers, hot-reload builds, and chat-GUI artifact previews live on
localhost/private addresses and change on every save — a 20-minute
cached copy would show a stale build exactly when freshness is the
point of fetching. The extract cache now declines loopback, private,
link-local, *.local, *.localhost, and single-label LAN hostnames on
both put and get. Hostname heuristics only (no DNS) — this is a
freshness carveout, not a security boundary; SSRF enforcement is
unchanged in tools/url_safety.py.

Public URLs keep the full TTL.
2026-08-25 04:21:45 -07:00
Teknium 8adef09be8 fix(web): extract cache serves only after policy + provider gates; rescue and format/provider isolation
Review fixes for #94618 (all three blockers reproduced by the reviewer
through the real web_extract_tool):

1. Cache lookup moved AFTER provider resolution and strict-selection
   validation, and gated per-URL on the website blocklist policy — a
   blocklist-blocked or misconfigured-backend call now behaves exactly
   as it would without a cache instead of serving cached content.
2. Rescue-served extract batches are never cached (mirrors the search
   memo's exclusion), keeping one-shot rescue one-shot.
3. Cache entries now get dedicated per-(url, format, provider) files
   instead of sharing the URL-keyed truncate-store file — html and
   markdown (or two backends') copies of one URL no longer overwrite
   each other, and switching extract backends within the TTL never
   serves the old backend's rendering.

Also from review: per-process index tmp filename (cross-process writers
can no longer truncate each other mid-write) and held flight locks are
never evicted from the bounded lock table (eviction could have allowed
a duplicate paid request).

New regression tests for formats/provider keying; E2E harness extended
with policy-block, strict-selection, rescue-two-call, and dual-format
scenarios — 6/6 pass; original 13/13 still pass.
2026-08-25 04:21:45 -07:00
Teknium 04603fc040 feat(web): TTL result caching for web_search + web_extract
Repeat searches (same normalized query + provider) within a 20-minute
TTL are served from an in-process memo, and concurrent identical
queries are single-flighted so a parallel subagent fan-out pays for
one vendor request instead of N. Requested limits bucket up to
10/20/50/100 so near-identical requests share an entry; callers get
their requested count sliced from the bucket.

Repeat extracts of the same URL are served from the existing
cache/web full-text store (previously written for read_file paging
but never read back), via a small JSON sidecar index. Disk-backed, so
CLI, gateway, cron, and subagents share it. Cached extracts re-run
the normal truncate pipeline, so per-call char_limit still works.

Both caches sit after every safety gate (secret-URL, SSRF, policy,
provider resolution) and directly around the paid vendor call — hits
skip only the network request. Only successful responses cache;
rescue-served responses are never cached (one-shot rescue must stay
one-shot). Config: web.cache_enabled (default on),
web.cache_ttl_minutes (default 20, clamped 1-1440).

Idea credit: query coalescing + num-bucketing pattern observed in
Apodex FrontierAgent (Apache-2.0).
2026-08-25 04:21:45 -07:00
Teknium 45b35f962f fix(mcp): tool-selection UIs stay in exclude mode instead of freezing include lists
Closes the two config-UI halves of the exclude-mode review (GottZ findings
5-7 on #94513):

- hermes mcp configure: on an exclude-mode server, unchecking a tool now
  APPENDS a literal exclude and re-checking drops it — glob patterns are
  preserved so future vendor tools keep getting filtered. Previously one
  uncheck converted the whole config to a frozen include list (globs
  silently deleted, new vendor tools invisible). Re-checked tools still
  shadowed by a kept glob get an explicit warning instead of a silent
  no-op.
- hermes tools MCP checklist: same exclude-mode write-back, plus display
  now matches excludes via matches_name_filter (fnmatch) — glob excludes
  previously rendered as if nothing were excluded.
- klaviyo manifest: post_install no longer tells users to append a param
  the URL already pins; now documents how to get the FULL surface.

Live-verified through the real cmd_mcp_configure path: exclude-mode server
with ['*_secret_*', 'docs'], uncheck beta + re-check docs ->
exclude becomes ['*_secret_*', 'beta'], no include written.
171 tests green across test_mcp_catalog/test_mcp_config/test_mcp_tool.
2026-08-25 04:21:37 -07:00
Jan-Stefan Janetzky 965689d13a fix(mcp-catalog): failed probe keeps the prior tool filter instead of wiping it
The probe-fail path ignored prior state entirely: with no manifest
default_enabled it wrote include=None, which pops the whole tools
block. For exclude-mode manifests default_enabled is necessarily unset,
and for the 30+ OAuth entries the entry rewrite precedes first auth —
so the common reinstall-while-unreachable case removed the curated
excludes and enabled every tool on next connect. The fallback order is
now: prior include > prior exclude > manifest default > no filter.
2026-08-25 04:21:37 -07:00
Jan-Stefan Janetzky 01177ed719 fix(mcp-catalog): don't announce a probe on the exclude-mode install path
The "Probing '<name>' for available tools..." line printed before the
exclude-mode short-circuit, which deliberately never probes (the test
suite asserts _probe_tools must not run there). Move the announcement
next to the actual probe call so install output matches what happened.
2026-08-25 04:21:37 -07:00
Jan-Stefan Janetzky fb1ec36a4b fix(mcp): treat tools.include: [] as an explicit empty whitelist
_normalize_name_filter([]) returns an empty set, which is falsy, so
_should_register fell through to "no filter" and registered every tool
— the exact opposite of what _apply_tool_selection wrote when the user
unchecked everything in the install checklist ("contributes nothing
until reconfigured"). Whitelist mode is now keyed on the include key
holding a valid filter shape (str/list/tuple/set) rather than on set
truthiness, at both the live-discovery and cached-manifest sites.
Invalid include values keep the old warn-and-ignore behaviour.
2026-08-25 04:21:37 -07:00
Jan-Stefan Janetzky 164e25a936 test(mcp-catalog): actually exercise the exfil-shaped-manifest rejection
The test wrote the evil manifest and imported install_entry but never
called it and asserted nothing, so the security gate in
_save_mcp_server/validate_mcp_server_entry was left uncovered. Now it
calls install_entry, expects CatalogError, and verifies the entry was
not persisted. Also drops the hardcoded ~/.hermes/ path from the
fixture args (AGENTS.md tests rule); the egress + exfil-hint shape
still trips both patterns.
2026-08-25 04:21:37 -07:00
Teknium 054cba271e fix(mcp): review findings — reinstall no longer clobbers user exclude lists + 4 curation gaps
Review blockers (independent reviewer on #94513):
1. Reinstalling an exclude-mode catalog entry wiped the user's edited
   tools.exclude, replacing it with manifest defaults. install_entry now
   reads the prior exclude (like it already did for include) and re-writes
   it verbatim on reinstall. Regression test added + sabotage-verified
   (fails on old behavior); include-priority test added too.
2. aws-knowledge: exclude aws___retrieve_skill — vendor SKILL.md loader is
   a vendor skill layer (live tools/list confirmed the tool exists).
3. betterstack: exclude list rewritten to cover the snake_case wire names
   (vendor's own header examples show remove_dashboard) via globs alongside
   the doc display-labels; caveat documented in the manifest — server is
   OAuth-gated so pre-auth enumeration is impossible.
4. railway: exclude railway-agent (opaque server-side agent delegation,
   acts outside Hermes's per-tool approval loop).
5. twelve-data: exclude oauth plumbing pseudo-tools + quota probe.
6. betterstack post_install no longer claims a fully-checked checklist —
   exclude-mode bypasses the checklist; text now describes the applied
   exclude list.

Live E2E: fresh temp HERMES_HOME — install applies manifest excludes,
user edit survives reinstall. 33/33 catalog tests green.
2026-08-25 04:21:37 -07:00
Teknium 88369af1c7 refine(mcp): debloat the catalog batch — vendor-doc tool audit applied to every entry
Policy applied (per Teknium direction, matching the Cloudflare precedent):
raw tool surfaces only — no server-side code-mode/search-execute layers, no
vendor tool_search; bloat (telemetry, feedback, docs-lookup, static-guidance
pseudo-tools, plan-gated upsells, dupe batch/compat shims) pruned via
manifest defaults.

DROPPED (meta-tool gateway IS the server, no vendor off-switch):
zapier (discover/enable/execute over 40k actions), wix (CallWixSiteAPI
generic invoke), customer-io (cio_read/write/delete_api generic HTTP
executors), omnisend (4 generic verb executors), apify (dynamic
actor-mount + telemetry-on-by-default), ramp (undocumented SQL/ETL layer,
no tool list, money-moving approval tools)

URL-LEVEL DEBLOAT (vendor-documented switches):
postman -> /minimal variant; klaviyo -> ?core-tools-only=true&
disable-tools-with-user-generated-content=true (262 -> ~40 tools)

CURATED default_excluded (20 entries) / default_enabled (kiwi, motherduck):
monday (GraphQL escape hatch trio...), close (voice-agent cluster that
places real AI phone calls, search/fetch layer, 14 excl), betterstack
(Execute query SQL hatch, 8 instruction pseudo-tools, team mgmt, 14 excl),
mixpanel (6 guidance pseudo-tools, bulk dupes, 10 excl), neon (search/
fetch, docs pair, logs beta, auth product, 10 excl), miro (6 deprecated),
gamma (viewer-tracking analytics), robinhood (upsell+social), dropbox,
todoist, fireflies, calendly, plaid, attio, gitlab, circleci, buildkite
(secrets-exposing get_job_env), semgrep, globalping, prisma-postgres,
motherduck (9-tool core enable), kiwi (feedback tool pruned)

Clean after audit (no changes needed): canva, clickup, linear-class lean
servers, twelve-data (read-only), algolia (read-only, vendor-curated),
indeed, strava (vendor read-only, no tool list published), craft,
wordpress-com (user-side toggles documented), trivago, alltrails,
deepwiki, context7, microsoft-learn, aws-knowledge, wolfram, twilio-docs
2026-08-25 04:21:37 -07:00
Teknium 9a37325717 feat(mcp): add 18 more live-verified remote MCPs from the final sweep
OAuth+DCR (14): mixpanel, algolia, klaviyo, amplitude, gitlab, circleci,
customer-io, omnisend, motherduck, strava, gamma, craft, wordpress-com,
robinhood (trade-execution caution in post_install)
No-auth (4): kiwi, trivago, alltrails, twilio-docs (search/read-only;
kiwi+trivago link out for booking, no in-conversation payment)

Probed and rejected: expedia, booking, uber, uber-eats, doordash,
instacart, spotify, audible, resy, stubhub, lastminute, coinbase,
posthog, dbt (all OAUTH_NO_DCR on the wire despite directory claims of
DCR); tripadvisor (token-in-URL endpoint); binance, crypto-com,
cash-app (policy: no in-chat trading/checkout beyond robinhood),
autodesk-help (too niche), viator (alltrails preferred)
2026-08-25 04:21:37 -07:00
Teknium 99f5aec733 feat(mcp): add Better Stack and Railway remote MCPs (OAuth+DCR, live-verified) 2026-08-25 04:21:37 -07:00
Teknium 3e570dd42d refine(mcp): drop github, perplexity, telnyx, exa, parallel-search from the catalog batch
- github: 44 tools / ~31K schema tokens duplicating the built-in gh CLI skill
- perplexity/telnyx: api_key friction — batch policy is browser-OAuth or no-auth only
- exa/parallel-search: redundant with Hermes's native web_search backend
2026-08-25 04:21:37 -07:00
Teknium 753f362a23 feat(mcp): add 34 official vendor-hosted remote MCP servers to the catalog
Competitor-parity expansion (Perplexity Computer, Manus, Claude, ChatGPT
connector catalogs) limited to endpoints a generic MCP client can actually
use. Every entry was live-probed: initialize handshake for no-auth servers,
RFC 9728 protected-resource metadata -> AS metadata with a
registration_endpoint for OAuth servers (true DCR only).

No-auth (7): deepwiki, context7, microsoft-learn, exa, aws-knowledge,
wolfram, telnyx*
API key (3): github (PAT bearer - GitHub reserves browser OAuth for
pre-registered IDEs), perplexity, telnyx
OAuth 2.1 + DCR (24): canva, zapier, monday, clickup, todoist, dropbox,
wix, miro, calendly, ramp, plaid, fireflies, neon, prisma-postgres,
postman, globalping, buildkite, semgrep, attio, close, apify, cloudinary,
parallel-search, twelve-data, indeed

Deliberately excluded after probing OAUTH_NO_DCR (vendor requires a
pre-registered app; generic clients cannot connect): Box, Slack, HubSpot,
Zoom, PagerDuty, Brex, Docusign, Stack Overflow, Google Workspace.
2026-08-25 04:21:37 -07:00
Teknium 3a1a3a1c8f feat(mcp): curated exclude list for cloudflare + glob tool filters + default_excluded manifests
The cloudflare entry's 3,320-endpoint surface is ~43% product families a
personal/dev account never touches (Zero Trust org-fleet suite, Magic
Transit/WAN, Cloudforce One, Radar analytics, API Shield, legacy
migration surfaces). Ship a 34-pattern curated exclude list in the
manifest: 3,320 -> 1,905 tools kept, and everything Cloudflare adds
later stays enabled by default.

Mechanism, two small extensions:
- tools/mcp_tool.py: tools.include/exclude entries containing glob
  metacharacters now match via fnmatch (plain names stay exact-match),
  so a product family is one pattern instead of hundreds of stale
  literals.
- hermes_cli/mcp_catalog.py: manifests may declare
  tools.default_excluded (mutually exclusive with default_enabled);
  install writes it to tools.exclude and skips the probe/checklist —
  a 3,320-row curses checklist is not a UX. Prior user include
  selections still win on reinstall.

Verified by replaying the real filter functions over the live-probed
3,320-tool list: 1,415 excluded, zero overmatch against a per-product
target audit; DNS/Workers/R2/D1/tunnels/Access/AI kept.
2026-08-25 04:21:37 -07:00
Teknium 53015d3eb5 feat(mcp): pin ?codemode=false so tool_search sees the full endpoint surface
The server's default Code Mode surface (search/execute meta-tools) is
itself a tool-discovery layer; stacking it under Hermes tool_search
would mean two search hops and an opaque 2-tool surface. With
?codemode=false each of the ~3,300 API endpoints registers as its own
tool with a full JSON Schema, and Hermes's own progressive disclosure
defers and searches the complete catalog — one layer, total
information. Verified live: tools/list returns 3,320 tools, all with
input schemas. post_install documents the trade-off and how to opt
back into Code Mode.
2026-08-25 04:21:37 -07:00
Teknium 90fd9a838b feat(mcp): add Cloudflare's official API MCP server to the catalog
Adds optional-mcps/cloudflare — Cloudflare's managed remote MCP server
(mcp.cloudflare.com/mcp) fronting the entire Cloudflare API (2,500+
endpoints across DNS, Workers, R2, KV, D1, Zero Trust, WAF, Pages)
through two Code Mode tools, search() and execute(), at a fixed ~1k-token
schema footprint. HTTP transport + native MCP OAuth 2.1 with DCR — no
install block, nothing to pin. post_install documents the scoped OAuth
grant, the bearer-token path for headless/CI, and Cloudflare's
product-specific servers for narrower surfaces.

Docs: mention Cloudflare in the hosted-OAuth MCP examples.
2026-08-25 04:21:37 -07:00
Teknium d736f5d53f fix(docker): digest-suffix shared-container identity labels so distinct keys never collide
Review finding on #94633: _sanitize_label_value is lossy ('team/workspace'
and 'team_workspace' both sanitize to 'team_workspace'; >63-char keys
truncate identically), and container reuse is label-keyed — so two teams
with DIFFERENT shared keys could silently attach to one running container
(filesystem, processes, env) while their host sandboxes stayed separate.
Shared-key labels now carry a sha256 digest suffix of the raw key
(deterministic across processes; plain profile labels unchanged for
backward compat). Docs also state the first-creator-wins rule for image/
mounts on a shared container. Adds adversarial collision tests.
2026-08-25 04:00:27 -07:00
Teknium 82b32f32ef feat(terminal): wire shared-container key into profile-scoped resolver and MEDIA delivery
Follow-up on @fangliquanflq's opt-in (#84775): after the profile-scoping fix
(#94560) the container cache key is resolved in _resolve_container_task_id,
so the shared key must unify profiles there too — 'shared:<key>' for every
session of every opted-in profile AND for CLI/no-session runs. Delivery adds
the shared sandbox layout as the first translation candidate. Empty key
keeps strict per-profile isolation; SSH ignores the key entirely.
2026-08-25 04:00:27 -07:00
fangliquanflq f5200a4c10 fix(config): declare shared Docker container key 2026-08-25 04:00:27 -07:00
fangliquanflq 7a67bd07a7 feat(docker): support shared container identities 2026-08-25 04:00:27 -07:00
Teknium 1ee524f77d fix(memory): bind the checkpoint gate to post-turn micro-compaction too
Independent review caught a compaction authority the gate missed:
post-turn micro-compaction (turn_finalizer -> _micro_compact) absorbs the
oldest exchanges into a rolling summary with no pre-compress checkpoint
hook in its path, and both compression.checkpoint_required and
compression.micro_compact could be enabled together — assistant evidence
could vanish into a summary the checkpoint filter later excludes, without
ever reaching the durable provider.

- agent_init: checkpoint_required forces micro-compaction off (warned),
  mirroring the native-compaction suppression
- turn_finalizer: defense-in-depth guard at the call site (attribute is
  plain mutable state a future path could flip on a live agent)
- behavioral regression test with a sabotage control (gate off proves the
  harness reaches the call site; gate armed proves zero calls)
- docs + config example mention the suppression; stale v1 test header fixed
2026-08-25 03:55:55 -07:00
Teknium 9e551d2931 refactor(memory): renumber checkpoint API — v1 is the implicit historical contract, v2 opts into fail-closed checkpoints
Per review: existing providers should not be retroactively re-versioned or
handed a changed payload. Version 1 is now the implicit historical
on_pre_compress() contract (best-effort, raw message list) that every
pre-existing provider is already on; the fail-closed checkpoint contract
becomes version 2. MemoryManager routes the raw transcript to v1 providers
unchanged and hands the host-normalized evidence list only to v2+ checkpoint
providers, so the plugin surface contract for shipped providers is
byte-identical with the gate off.
2026-08-25 03:55:55 -07:00
Jan-Stefan Janetzky 8cc379b528 fix(memory): bind the checkpoint gate to every compaction authority
The fail-closed gate lived only in compress_context(), but two native
lossy owners compact without ever crossing it (review on #93996):

- codex app-server: in "native"/"off" auto-compaction mode (native is the
  default) Hermes preflight is skipped and the codex agent compacts its
  own thread inside run_turn() — the compress_context() rejection was
  unreachable. init_agent now refuses checkpoint_required together with
  api_mode=codex_app_server (BLOCKED_MISSING_PREREQUISITE, extracted as a
  testable guard), and run_codex_app_server_turn() fails closed as
  defense in depth before a turn can reach the codex-owned boundary.
- Responses server-side native compaction:
  native_compaction_context_management() now returns None while the gate
  is armed, so context_management never goes on the wire and the
  checkpoint-aware Hermes compressor stays authoritative. The suppression
  is logged once per process, not silently applied.

Regressions: checkpoint_required + app-server raises before run_turn()
(the session is never created); checkpoint_required keeps
context_management off the wire while the plain configuration still
produces it; the init guard refuses exactly the incompatible pair. Docs
and cli-config.yaml.example describe both bindings.

Refs #93986
2026-08-25 03:55:55 -07:00
Jan-Stefan Janetzky 70d0b1fffb fix(memory): review follow-ups for the pre-compress checkpoint contract
Addresses the review on #93996:

- gateway: hygiene and manual /compress load the memory provider only when
  compression.checkpoint_required is enabled (skip_memory=not required).
  The historical fast path — no provider init, no best-effort hook — is
  back for everyone who did not opt in, so default behavior is truly
  unchanged.
- conversation_compression: assistant messages carrying both prose and
  tool_calls keep their prose in the checkpoint evidence (the tool_calls
  payload is stripped, the original message is not mutated); pure
  tool-call wrappers without prose are still dropped.
- tests: legacy-database regression proving the _compressed_summary column
  is added by the declarative _reconcile_columns() path on a plain reopen
  (no version-gated migration needed — append_message works right after),
  plus coverage for the prose-preserving filter.
- docs: providers must implement idempotent, content-keyed checkpoint
  writes — a fail-closed block means the next attempt re-runs
  on_pre_compress over largely the same transcript.

Refs #93986
2026-08-25 03:55:55 -07:00
Jan-Stefan Janetzky 3c31c44880 docs(memory): document the pre-compress checkpoint contract
Adds a 'Pre-Compress Checkpoints (fail-closed)' section to the memory
provider plugin guide: the versioned opt-in attribute, the operator-side
compression.checkpoint_required gate, fail-closed semantics, and the
normalized evidence contract including the persistent summary marker.

Refs #93986
2026-08-25 03:55:55 -07:00
Jan-Stefan Janetzky 1104ffe0b9 feat(memory): opt-in fail-closed pre-compress checkpoint contract (API v1)
Context compression is intentionally lossy. Deployments that archive
transcript evidence to an external durable store before compaction had no
way to guarantee the archive actually happened: MemoryManager.on_pre_compress
swallows provider failures by design, so a failed archive silently degraded
into data loss.

This adds an opt-in, provider-agnostic checkpoint contract:

- memory_provider: PRE_COMPRESS_CHECKPOINT_API_VERSION = 1; providers opt in
  by advertising pre_compress_checkpoint_api_version. Version 0 keeps the
  historical best-effort hook semantics.
- memory_manager: supports_pre_compress_checkpoint() capability probe;
  on_pre_compress(require_checkpoint=True) propagates checkpoint-provider
  failures and raises when no capable provider completed the checkpoint.
- conversation_compression: new compression.checkpoint_required config key
  (default false, documented in cli-config.yaml.example). When enabled,
  compaction fails closed with BLOCKED_MISSING_PREREQUISITE (the
  uncompressed transcript is preserved) unless a checkpoint-capable provider
  confirms the durable checkpoint. Providers receive normalized direct
  user/assistant evidence: tool rows, system messages, tool-call wrappers,
  and prior compaction summaries are filtered host-side into one stable
  contract. codex_app_server compaction is rejected under the gate because
  it exposes no truthful pre-compaction transcript boundary.
- hermes_state: persistent _compressed_summary column (declarative schema
  migration via _reconcile_columns) so summary provenance survives process
  restarts; only the resume model history carries the marker, keeping
  get_messages_as_conversation on its existing contract.
- gateway: the lossy hygiene/auto-compact paths load the memory provider
  (skip_memory=False) so a required checkpoint also guards those rewrites.

The gate arms only on an explicit boolean True (bare-MagicMock agents in
existing tests have truthy auto-attributes). Default behavior is unchanged:
checkpoint_required=false preserves best-effort semantics for all existing
providers. Contract tests, including a restart round-trip of the summary
marker, in tests/agent/test_pre_compress_checkpoint_contract.py.

Refs #93986
2026-08-25 03:55:55 -07:00
Teknium 76e306c458 refactor(tools): remove expired BFL FLUX 3 promo core tools (migration v39); FLUX 3 stays via video_gen/FAL for subscribers (#94599)
* refactor(tools): remove expired bfl_flux3_* promo tools; FLUX 3 rides the video_gen provider surface

* test: relay-cutover migration asserts >= v38, not the version literal
2026-08-25 02:45:10 -07:00
Teknium 0268c0b8c0 chore: map 2ndNatureAI attribution 2026-08-25 02:31:34 -07:00
Teknium ce9b9a6351 feat(computer_use): guide models from full-screen grabs to interactive lanes
Full-screen captures carry no element tree, so the CaptureResult now has a
'note' field surfaced in the tool summary telling the model to call
capture(app='<AppName>') or capture(app='desktop') when it needs to act on
what it sees. Schema description updated to distinguish app='screen'
(composited full-screen image) from app='desktop' (shell surface with
clickable elements); docs + regression tests (14, sabotage-verified) added.
2026-08-25 02:31:34 -07:00
2ndNatureAI aeac982223 fix(computer_use): route explicit screen capture to get_desktop_state
'Screenshot my screen' previously resolved the 'screen' sentinel to the OS
shell window (Progman/WorkerW) via list_windows — capturing the wallpaper +
icons layer, never the windows actually displayed. cua-driver's
get_desktop_state does a real composited full-screen grab; the
screen/fullscreen/all sentinels now route there directly, bypassing window
enumeration (which also keeps screenshots working when Windows UIA
enumeration hangs — trycua/cua#2110/#2113).

app='desktop' keeps the shell-window lane so desktop icons/taskbar stay
clickable.

Salvaged from PR #60081 by @2ndNatureAI (surgical reapply; original branch
predates the capture-routing refactor).
2026-08-25 02:31:34 -07:00
Teknium 15f7b7293c fix(terminal): persistent Docker containers are profile-scoped, not per-session
Commit a270c4ade's session-key fallback in _resolve_container_task_id was
added to stop cross-profile SSH environment reuse, but it wasn't backend-
gated: persistent Docker silently fragmented into one container per gateway
session, breaking the product contract (one long-lived container per profile,
shared by CLI and every session of that profile). #93950's vanishing MEDIA
attachments were downstream damage.

- persistent Docker (container_persistent: true) now keys to the profile:
  literal 'default' for the default profile (same container as CLI),
  'profile:<name>' for named profiles
- SSH and non-persistent Docker keep session scoping (the original leak fix
  and the #82731 isolation contract are untouched)
- gateway MEDIA translation follows the profile layout and keeps the legacy
  bug-window per-session sandboxes as fallback candidates, trying each until
  the file resolves — old sessions self-heal, no migration
- /root/.hermes credential-surface refusal preserved across all layouts
2026-08-25 02:30:38 -07:00
kshitij 4c1f53be10 Merge pull request #94568 from kshitijk4poor/fix/85125-2e-approval-outcome-parity-v2
fix(approval): machine-readable outcome parity on the gateway tails + sudo human-wait exclusion (#85125 2e)
2026-08-25 13:31:49 +05:30
kshitij dc3716d451 Merge pull request #94536 from kshitijk4poor/salvage/94439-computer-use-media-path
fix(gateway): widen computer-use media path repair to background and cron delivery
2026-08-25 13:31:05 +05:30
kshitijk4poor c8c3f4c448 fix(approval): machine-readable outcome parity on the gateway tails + sudo human-wait exclusion (#85125 2e) 2026-08-25 13:27:20 +05:30
kshitijk4poor d634b37047 refactor(gateway): retire private repair alias per replay_cleanup precedent
Phase 2c on the full final diff flagged the re-export shim as
contradicting the adjacent house pattern (agent.replay_cleanup import,
which documents retiring private aliases once tests migrate). Migrate
all six tests to the canonical gateway.media_repair seam, import the
canonical name in run.py, and drop the dead 'and result' guard at the
background-task call site.
2026-08-25 13:25:48 +05:30
kshitijk4poor e5032945cb chore: map contributor email for salvage attribution
Map macd@google.com (Mark McDonald, @markmcd) so the contributor
attribution check passes for the salvage of PR #94522.
2026-08-25 13:14:30 +05:30
Mark McDonald eaf6545ab4 docs(gemini): update to use latest gemini models 2026-08-25 13:14:30 +05:30
kshitijk4poor 105999a0c9 refactor(gateway): unify computer-use repair call sites after review
- Make repair_explicit_computer_use_media_paths fail-open internally
  (cosmetic repair must never abort delivery); drop the cron-only
  try/except so all three call sites are identical one-liners.
- Drop cron's redundant 'MEDIA:' pre-check (helper early-returns).
- Document the intentional lazy BasePlatformAdapter import (verified:
  no cycle either way; keeps module import cheap for cron processes).
- Point the two new regression tests at the canonical
  gateway.media_repair seam; pre-existing tests keep pinning the
  gateway.run re-export shim.
- Docstring: matching is case-insensitive, say so.
2026-08-25 13:06:43 +05:30
kshitijk4poor bb0d5503c2 fix(gateway): widen computer-use media path repair to sibling surfaces
Follow-up to the salvaged fix from PR #94439:

- Extract the repair into gateway/media_repair.py (shared module) and
  re-export under the historical private name in gateway/run.py.
- Wire the repair into the two bypassed delivery surfaces: gateway
  background tasks (_run_background_task_inner) and cron job delivery
  (cron/scheduler.py) — both call agent.run_conversation directly and
  never pass the main turn chokepoint.
- Fail closed on malformed/truncated JSON tool results: parse JSON-looking
  content first instead of regex-scanning the raw string, which yielded a
  doubled-backslash path artifact and rewrote the response to a path the
  model never wrote.
- Deduplicate the tool_name_by_call_id builder (three verbatim copies in
  gateway/run.py) into the shared module; hoist the abs-path prefix regex.
- Add regression tests: malformed-JSON fail-closed (mutation-checked) and
  the compression-fallback last-user slice (incl. no-user fail-closed).
2026-08-25 13:06:43 +05:30
Teknium 760d3af0d7 chore: map contributor email for salvage attribution 2026-08-25 00:31:14 -07:00
Guilherme Artiles 8ad20d065a test(curator): assert the instruction through the delivered prompt, not the source
The first version of this test used inspect.getsource() on skill_manager_tool
and regex-parsed the guarded action literals. AGENTS.md bans source-text tests,
and the ban is right here: that test would pass against a guard wired to the
wrong call site and fail on a pure rename, neither of which is the thing worth
guarding.

Replaced with a behavioral assertion in the shape of the neighbouring
dry-run-banner test: stub _run_llm_review, run run_curator_review, and assert
the prompt that actually reached the model names skill_view and all four
guarded actions (edit, patch, write_file, remove_file).

It still discriminates: with the prompt block removed, action=edit and
action=remove_file no longer appear anywhere in the assembled prompt (the
toolset list only mentions patch, create, write_file and delete), so the test
fails. Runtime guard behavior stays where it belongs, in
tests/tools/test_skill_manager_tool.py.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-25 00:31:14 -07:00
Guilherme Artiles 610e2e02fb fix(curator): tell the background reviewer to read before it writes
`_background_review_read_before_write_guard` refuses a background-review
`skill_manage` write when the target file was not loaded via `skill_view` in
the same review turn (patch, edit, write_file over an existing file,
remove_file).

`CURATOR_REVIEW_PROMPT` never says so. It lists `skill_view` only under "read
the current landscape", so the reviewer goes straight to the write and every
mutation is refused. The failure is silent from the outside: the curator run
completes, writes nothing, and reads like a pass that simply found nothing to
consolidate. On our deployment that was 32 of 32 attempted writes rejected over
48h before anyone read the logs.

This adds the missing instruction to the toolset block, plus a test that fails
if a future guarded action is added to `skill_manager_tool` without being named
in the prompt — the guard and the prompt have to drift together or not at all.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-25 00:31:14 -07:00
Teknium a70d2ffce5 fix(background-review): teach review prompts the enforced read-before-write handshake
The skill_manage guard (added in #55906) refuses any patch/edit of an
existing SKILL.md, or overwrite/removal of an existing support file,
unless the exact target was loaded via skill_view during the review.
Neither _SKILL_REVIEW_PROMPT nor _COMBINED_REVIEW_PROMPT ever mentioned
this, so models routinely issued the write without the pre-read, got
refused, and burned review iterations (#62397).

Both prompts now carry a Read-before-write section scoped to the
guard's actual contract: existing targets only, exact-path pre-read for
support files, transcript quotes don't count, new skills/new support
files exempt, and a bounded one-view-one-retry recovery instead of a
loop. Direction follows #60331 by @kkwills13 with the scope corrections
requested in review (existing-target-only wording, no delete claim,
bounded retry, contract tests for both prompt variants).

Fixes #62397.
2026-08-25 00:18:35 -07:00
kshitijk4poor b0cf2597c2 fix: follow-up for salvaged PR #93985 — cache key, snapshot, dead code
- Key _user_space_cache on _conn_snapshot instead of client object identity,
  so _new_client() results from the same connection share the cached user
  (previously every on_memory_write triggered an uncached /api/v1/system/status
  probe with a 30s default timeout)
- Thread a short timeout (0.05s) through the write-path identity probe
- Harden _tool_remember to snapshot the client before URI construction + POST,
  matching the pattern already established in on_memory_write
- Remove dead instance method _user_scoped_uri (zero callers; all call sites
  use the module-level function directly)

Co-authored-by: ehz0ah <haozhe4547@gmail.com>
2026-08-25 12:27:21 +05:30
ehz0ah 4387e03960 fix(memory): keep OpenViking identity operations consistent 2026-08-25 12:27:21 +05:30