Skip to content

fix(detect): update Grok agent detection rules for Grok Build 1.0.x #2764

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
jasonchen36 wants to merge 1 commit into from
Closed

fix(detect): update Grok agent detection rules for Grok Build 1.0.x #2764

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 1 addition & 7 deletions scripts/agent_detection_manifest_check.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,13 +56,7 @@
# Keep engine-2 clients on the OSC-capable manifest until an engine-3 release
# can consume top_non_empty_lines. Remove this entry when the website publishes
# the bundled Grok manifest.
STAGED_WEBSITE_MANIFESTS = {
"grok": (
"2026.07.16.2",
"2026.07.16.1",
"1f35b3271a96cf830c64bed78751619bfd8013c277c0d7c0f999b7a433895f28",
),
}
STAGED_WEBSITE_MANIFESTS = {}


def parse_args() -> argparse.Namespace:
Expand Down
2 changes: 2 additions & 0 deletions scripts/test_agent_detection_manifest_check.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -73,13 +73,15 @@ def test_rejects_website_version_lower_than_bundled(self):
with self.assertRaisesRegex(check.CheckError, "lower than bundled"):
check.validate_catalog(website, bundled_manifests, engine_version=1)

@unittest.skipUnless(check.STAGED_WEBSITE_MANIFESTS, "no staged website manifests configured")
def test_allows_explicitly_staged_website_manifest(self):
with tempfile.TemporaryDirectory() as tmp:
bundled, website = staged_grok_dirs(Path(tmp))

bundled_manifests = check.load_manifest_dir(bundled, engine_version=3)
check.validate_catalog(website, bundled_manifests, engine_version=3)

@unittest.skipUnless(check.STAGED_WEBSITE_MANIFESTS, "no staged website manifests configured")
def test_rejects_mutated_staged_website_manifest(self):
with tempfile.TemporaryDirectory() as tmp:
bundled, website = staged_grok_dirs(Path(tmp))
Expand Down
113 changes: 75 additions & 38 deletions src/detect/manifests/grok.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@
id = "grok"
version = "2026.07.16.2"
min_engine_version = 3
updated_at = "2026-07-16T00:00:00Z"
version = "2026.08.13.1"
min_engine_version = 2
updated_at = "2026-08-13T00:00:00Z"
aliases = ["grok-build"]

# Evidence: Grok Build 0.2.101 source and live pane reads.
# Evidence: Grok Build 1.0.3 source and live pane reads.
#
# Grok emits OSC 0 titles by default. Idle is "grok" or
# "<session> - grok". During a turn, the configured title gains a braille
Expand All @@ -25,9 +25,17 @@ aliases = ["grok-build"]
# "┃ z (○) Type your answer here"
# with footer hints "1/3:select │ Ctrl+o:yolo │ Ctrl+c:cancel" (permission)
# or "Esc:unselect │ Tab:scrollback │ Shift+x:dismiss" (question dialog).
# Idle footers end with "Ctrl+.:shortcuts" and never contain "Esc:cancel".
# The startup splash draws its logo with braille characters, so working
# rules must anchor on the [stop] chip, not on a bare spinner glyph.
# Idle footers end with "Ctrl+x:shortcuts" or "Ctrl+.:shortcuts" and never contain "Esc:cancel".
#
# Background subagent hangover (Grok Build):
# After the parent turn yields, OSC title/progress and the footer can go
# idle while a child tool/subagent is still running. Live chrome then looks
# like:
# "┃ ◆ Run Implement Phases J–O harness modules [hooks: 3]"
# "Worked for 11m13s. 1 command still running."
# Completed tools drop the ┃ gutter; option dialogs use "┃ N (○)" instead
# of "┃ ◆". These working rules must outrank osc_title_idle (1140) and
# osc_progress_idle (850), or the pane false-reports done/idle mid-wait.

[[rules]]
id = "osc_title_blocked"
Expand Down Expand Up @@ -74,51 +82,91 @@ any = [
{ contains = ["←/→:scope"] },
]

# Grok clears its OSC busy signals while background work runs. The first
# non-empty row is pinned application chrome, where this animated chip shows
# the number of running background tasks and disappears when the count is zero.
# Idle prompt footer rule outranks historical transcript lines in scrollback.
[[rules]]
id = "prompt_hints_idle"
state = "idle"
priority = 1160
region = "bottom_non_empty_lines(2)"
visible_idle = true
any = [
{ contains = ["ctrl+.:shortcuts"] },
{ contains = ["ctrl+x:shortcuts"] },
{ contains = ["shift+tab:mode"] },
]
not = [
{ contains = ["esc:cancel"] },
{ contains = ["ctrl+c:cancel"] },
]

# Parent turn has yielded, but a background tool/subagent is still live.
[[rules]]
id = "background_work_chip_working"
id = "background_commands_still_running"
state = "working"
priority = 1170
region = "top_non_empty_lines(1)"
priority = 1150
region = "whole_recent"
visible_working = true
line_regex = ['[⋅:⸬⁙.·]\s+[1-9][0-9]*\s+│']
regex = ['\b\d+\s+commands?\s+still\s+running\b']

[[rules]]
id = "osc_progress_working"
id = "active_subagent_gutter_working"
state = "working"
priority = 1150
region = "osc_progress"
priority = 1145
region = "whole_recent"
visible_working = true
regex = ['^4;1;-1$']
# Active tool/subagent rows keep the ┃ gutter + ◆ marker.
# Permission option rows use "┃ N (○)" / "┃ z (○)" instead.
line_regex = ['^\s*┃\s+◆\s+\S']

[[rules]]
id = "waiting_on_subagent_working"
state = "working"
priority = 1142
region = "whole_recent"
visible_working = true
any = [
{ contains = ["waiting on subagent"] },
{ contains = ["waiting for subagent"] },
{ regex = ['\bwaiting on \d+ (?:background )?(?:task|command|subagent)s?\b'] },
]

[[rules]]
id = "osc_title_idle"
state = "idle"
priority = 1100
priority = 1140
region = "osc_title"
visible_idle = true
regex = ['(?:^| - )grok$']
regex = ['(?i)(?:^|.*[\s\-_—–])grok(?:[\s\-_—–].*)?\s*$']
not = [
{ regex = ['[\x{2800}-\x{28FF}]'] },
{ contains = ["[stop]"] },
]

# After known idle titles are excluded, any other non-empty Grok title is
# active. OSC progress remains authoritative when a custom title omits the
# spinner but the terminal supports progress reporting.
# Active title with spinner or [stop] chip.
[[rules]]
id = "osc_title_working"
state = "working"
priority = 1000
region = "osc_title"
visible_working = true
regex = ['\S']
any = [
{ regex = ['[\x{2800}-\x{28FF}]'] },
{ contains = ["[stop]"] },
]

# Deprioritized below prompt/title idle rules because Grok 1.0.x does not clear 9;4 progress on yield.
[[rules]]
id = "osc_progress_working"
state = "working"
priority = 900
region = "osc_progress"
visible_working = true
regex = ['^4;1;-1$']

[[rules]]
id = "osc_progress_idle"
state = "idle"
priority = 950
priority = 850
region = "osc_progress"
visible_idle = true
regex = ['^4;0;0$']
Expand All @@ -129,6 +177,7 @@ state = "working"
priority = 200
region = "whole_recent"
visible_working = true
# Allow optional [↓]/chips between the status text and [stop].
line_regex = ['^\s*[\x{2801}-\x{28FF}]\s.*\[stop\]\s*$']

[[rules]]
Expand All @@ -137,7 +186,7 @@ state = "working"
priority = 190
region = "bottom_non_empty_lines(2)"
visible_working = true
contains = ["esc:cancel", "ctrl+.:shortcuts"]
contains = ["esc:cancel"]

# Pre-0.2.x working chrome kept for older Grok Build releases.
[[rules]]
Expand All @@ -150,15 +199,3 @@ any = [
{ all = [{ contains = ["ctrl+c:cancel", "ctrl+enter:interject"] }, { contains = ["waiting"] }] },
{ line_regex = ['^\s*[\x{2801}-\x{28FF}]\s+(Run|Read|Search|List)\b'] },
]

[[rules]]
id = "prompt_hints_idle"
state = "idle"
priority = 100
region = "bottom_non_empty_lines(2)"
visible_idle = true
contains = ["ctrl+.:shortcuts"]
not = [
{ contains = ["esc:cancel"] },
{ contains = ["ctrl+c:cancel"] },
]
106 changes: 77 additions & 29 deletions website/agent-detection/grok.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@
id = "grok"
version = "2026.07.16.1"
version = "2026.08.13.1"
min_engine_version = 2
updated_at = "2026-07-16T00:00:00Z"
updated_at = "2026-08-13T00:00:00Z"
aliases = ["grok-build"]

# Evidence: Grok Build 0.2.101 source and live pane reads.
# Evidence: Grok Build 1.0.3 source and live pane reads.
#
# Grok emits OSC 0 titles by default. Idle is "grok" or
# "<session> - grok". During a turn, the configured title gains a braille
Expand All @@ -25,9 +25,17 @@ aliases = ["grok-build"]
# "┃ z (○) Type your answer here"
# with footer hints "1/3:select │ Ctrl+o:yolo │ Ctrl+c:cancel" (permission)
# or "Esc:unselect │ Tab:scrollback │ Shift+x:dismiss" (question dialog).
# Idle footers end with "Ctrl+.:shortcuts" and never contain "Esc:cancel".
# The startup splash draws its logo with braille characters, so working
# rules must anchor on the [stop] chip, not on a bare spinner glyph.
# Idle footers end with "Ctrl+x:shortcuts" or "Ctrl+.:shortcuts" and never contain "Esc:cancel".
#
# Background subagent hangover (Grok Build):
# After the parent turn yields, OSC title/progress and the footer can go
# idle while a child tool/subagent is still running. Live chrome then looks
# like:
# "┃ ◆ Run Implement Phases J–O harness modules [hooks: 3]"
# "Worked for 11m13s. 1 command still running."
# Completed tools drop the ┃ gutter; option dialogs use "┃ N (○)" instead
# of "┃ ◆". These working rules must outrank osc_title_idle (1140) and
# osc_progress_idle (850), or the pane false-reports done/idle mid-wait.

[[rules]]
id = "osc_title_blocked"
Expand Down Expand Up @@ -74,40 +82,91 @@ any = [
{ contains = ["←/→:scope"] },
]

# Idle prompt footer rule outranks historical transcript lines in scrollback.
[[rules]]
id = "osc_progress_working"
id = "prompt_hints_idle"
state = "idle"
priority = 1160
region = "bottom_non_empty_lines(2)"
visible_idle = true
any = [
{ contains = ["ctrl+.:shortcuts"] },
{ contains = ["ctrl+x:shortcuts"] },
{ contains = ["shift+tab:mode"] },
]
not = [
{ contains = ["esc:cancel"] },
{ contains = ["ctrl+c:cancel"] },
]

# Parent turn has yielded, but a background tool/subagent is still live.
[[rules]]
id = "background_commands_still_running"
state = "working"
priority = 1150
region = "osc_progress"
region = "whole_recent"
visible_working = true
regex = ['^4;1;-1$']
regex = ['\b\d+\s+commands?\s+still\s+running\b']

[[rules]]
id = "active_subagent_gutter_working"
state = "working"
priority = 1145
region = "whole_recent"
visible_working = true
# Active tool/subagent rows keep the ┃ gutter + ◆ marker.
# Permission option rows use "┃ N (○)" / "┃ z (○)" instead.
line_regex = ['^\s*┃\s+◆\s+\S']

[[rules]]
id = "waiting_on_subagent_working"
state = "working"
priority = 1142
region = "whole_recent"
visible_working = true
any = [
{ contains = ["waiting on subagent"] },
{ contains = ["waiting for subagent"] },
{ regex = ['\bwaiting on \d+ (?:background )?(?:task|command|subagent)s?\b'] },
]

[[rules]]
id = "osc_title_idle"
state = "idle"
priority = 1100
priority = 1140
region = "osc_title"
visible_idle = true
regex = ['(?:^| - )grok$']
regex = ['(?i)(?:^|.*[\s\-_—–])grok(?:[\s\-_—–].*)?\s*$']
not = [
{ regex = ['[\x{2800}-\x{28FF}]'] },
{ contains = ["[stop]"] },
]

# After known idle titles are excluded, any other non-empty Grok title is
# active. OSC progress remains authoritative when a custom title omits the
# spinner but the terminal supports progress reporting.
# Active title with spinner or [stop] chip.
[[rules]]
id = "osc_title_working"
state = "working"
priority = 1000
region = "osc_title"
visible_working = true
regex = ['\S']
any = [
{ regex = ['[\x{2800}-\x{28FF}]'] },
{ contains = ["[stop]"] },
]

# Deprioritized below prompt/title idle rules because Grok 1.0.x does not clear 9;4 progress on yield.
[[rules]]
id = "osc_progress_working"
state = "working"
priority = 900
region = "osc_progress"
visible_working = true
regex = ['^4;1;-1$']

[[rules]]
id = "osc_progress_idle"
state = "idle"
priority = 950
priority = 850
region = "osc_progress"
visible_idle = true
regex = ['^4;0;0$']
Expand All @@ -118,6 +177,7 @@ state = "working"
priority = 200
region = "whole_recent"
visible_working = true
# Allow optional [↓]/chips between the status text and [stop].
line_regex = ['^\s*[\x{2801}-\x{28FF}]\s.*\[stop\]\s*$']

[[rules]]
Expand All @@ -126,7 +186,7 @@ state = "working"
priority = 190
region = "bottom_non_empty_lines(2)"
visible_working = true
contains = ["esc:cancel", "ctrl+.:shortcuts"]
contains = ["esc:cancel"]

# Pre-0.2.x working chrome kept for older Grok Build releases.
[[rules]]
Expand All @@ -139,15 +199,3 @@ any = [
{ all = [{ contains = ["ctrl+c:cancel", "ctrl+enter:interject"] }, { contains = ["waiting"] }] },
{ line_regex = ['^\s*[\x{2801}-\x{28FF}]\s+(Run|Read|Search|List)\b'] },
]

[[rules]]
id = "prompt_hints_idle"
state = "idle"
priority = 100
region = "bottom_non_empty_lines(2)"
visible_idle = true
contains = ["ctrl+.:shortcuts"]
not = [
{ contains = ["esc:cancel"] },
{ contains = ["ctrl+c:cancel"] },
]
Loading