# IDs and limits - Obss

Product reference
IDs and limits
Look up what each Obss ID prefix names and where you see it, and the product limits for paging, tags, schedules, launches, and runs.
ID format
Most public IDs are a prefix that names the kind of record, an underscore, and 24 lowercase hexadecimal characters, such as
exp_4c19a7e2b05d3f8a61c2e9d0
. An ID never changes, even when you rename the record. Pass IDs exactly as Obss returns them; these docs shorten them for readability, as in
exp_8f2c41d0
.
Record IDs
Prefix
Names
Where you see it
prj_
A project
Its dashboard address (
/projects/prj_…
),
obss projects list
,
obss projects current
, and
--project-id
tsk_
A task, in every version
Task pages,
obss catalog list
, and
--task-id
tskver_
One version of a task
Catalog and run results, and
--task-version-id
fld_
A file
Manage → Files
, the catalog, and
--file-id
env_
A machine
Manage → Machines (Environments)
, the catalog, and
--machine-id
skl_
A skill
Manage → Tools
, the catalog, and
--tool-id
mcp_
An MCP server
Manage → Tools
, the catalog,
obss mcp accounts
, and
--tool-id
cli_
A CLI
Manage → Tools
, the catalog, and
--tool-id
sdk_
An SDK
Manage → Tools
, the catalog, and
--tool-id
sklver_
,
mcpver_
,
cliver_
,
sdkver_
One version of a skill, MCP server, CLI, or SDK
Catalog results, and
--tool-version-id
trt_
A treatment (toolset)
Manage → Treatments (Toolsets)
, the catalog, and
--treatment-id
agm_
One agent with one model
obss catalog list --kind agent
, launch plans, and
--agent-model-id
exp_
An experiment
Experiment pages,
obss experiments list
, and
--experiment-id
sch_
A schedule
Schedule pages,
obss schedules list
, and
--schedule-id
run_
A run. Each trial is a run with its own
run_
ID.
obss runs list
, friction evidence, Evals and Insights results, and
--run-id
frc_
A friction
Friction pages,
obss frictions list
, and
--friction-id
cmp_
One comparison in an experiment, kept when Evals run again
obss evals compare
,
--comparison
, and
comparison_id
brf_
An experiment’s written Brief
obss experiments brief
tag_
A tag
obss tags list
aic_
A model provider connection
obss project ai inspect --view providers
aim_
A model on a provider connection
obss project ai inspect --view models
, and Evals and Insights model choices
upl_
An archive upload
obss uploads create
and
obss uploads get
qtx_
One run balance transaction
obss quota ledger
chat_
One Ask Obss chat
Ask Obss, including editor chats through
obss acp
Run scope flags
take
exp_
,
sch_
,
tsk_
,
tskver_
,
trt_
,
skl_
,
mcp_
,
cli_
,
sdk_
,
sklver_
,
mcpver_
,
cliver_
,
sdkver_
,
fld_
,
env_
,
agm_
, and
frc_
.
Tags
go on
tsk_
,
fld_
,
env_
,
skl_
,
mcp_
,
cli_
,
sdk_
,
trt_
, and
exp_
.
Operation IDs
Obss returns these for one operation, for you to send back. Their length varies.
Prefix
Names
Where you see it
adm_
A launch
obss experiments launch
,
obss experiments status
and
watch
, and Test agent runs
del_
A deletion
obss resources delete
and
obss deletions status
eval_
An Evals attempt
obss evals list
and
obss evals get
insights_
An Insights attempt
obss insights list
and
obss insights get
A public experiment link has a
pub_
ID with 32 hexadecimal characters, shown by
obss experiments publications list
and on
Share results
.
Keys
Some records have a key instead of an ID.
Key
Names
Example
Agent key
An agent with any model: a run’s
agentId
, and what
--agent-id
takes
claude-code
Custom agent key
A
custom agent
acp-0123456789abcdef
Task key, skill key
A task or skill by its key, for filters such as
--task-key
checkout
Agent keys use lowercase letters, digits, and hyphens, never an underscore, so they never look like public IDs. Send a cursor (
nextCursor
) back exactly as you received it. An attempt ID (
--attempt-id
) is a string you choose; reuse it only to retry the same change (
CLI conventions
).
Paging limits
A
--limit
(or the plugin’s
limit
) above the maximum is capped, not refused. Responses are capped at 20 KiB, so a page can hold fewer items. Continue with
nextCursor
while
hasMore
is true.
Command
Plugin (MCP) tool
Default
Maximum
obss catalog list
inspect_catalog
100
500
obss resources inspect
inspect_resource
100
500
obss experiments list
list_experiments
50
200
obss experiments publications list
list_experiment_publications
50
200
obss gallery get
get_public_experiment
50
200
obss schedules list
list_schedules
50
200
obss runs list
list_runs
50
200
obss runs trajectory
get_run_trajectory
100 steps
500 steps
obss frictions list
list_frictions
25
100
obss evals list
,
obss insights list
list_evals
,
list_insights
25
100
obss evals get
,
obss insights get
get_evals_results
,
get_insights_results
50
200
obss results analyze
analyze_results
100 ranked items, not paged
obss evals compare
get_evals_comparisons
200
200
obss project ai inspect
inspect_project_ai
100
500
obss quota ledger
list_quota_ledger
50
100
obss notifications list
list_notifications
25
50
obss notifications acknowledge --all
acknowledge_notifications
500
500
obss data query
query_data
200 rows
obss github browse
browse_github_source
Repositories 10, refs 100, paths 100
Repositories 30, refs 200, paths 1,000
Comparisons page with the same
--kind
and
--label
(
kind
and
label
in the plugin), and a cursor is refused once the comparisons change.
Content and wait limits
Limit
Value
Where it applies
Bytes in one content page
12,288
obss runs transcript
,
obss runs output
, the byte views of
obss evals get
and
obss insights get
, and
obss resources inspect
and
archive
. Page by offset until
eof
.
Wait for a change (
--wait-ms
)
25,000 ms
obss experiments status
,
obss evals get
, and
obss insights get
. In the plugin,
wait_ms
on
get_launch_status
,
get_experiment_status
,
get_evals_results
, and
get_insights_results
.
Runs analyzed together (
--runs
)
3
obss results analyze
Product limits
Limit
Value
Details
Runs in one
Compare
3
Read results
Matrix cells (tasks × combinations) that
Comparisons
reads
10,000. Larger returns
experiment_too_large
(
Too large to compare
).
Comparisons
Agents that
Comparisons
compares with each other
6. With more, agents aren’t compared with each other, and there’s no
Does a treatment help one agent more?
Comparisons
Differences in a
Brief
5, each with up to 3 frictions, and up to 5 other frictions
Brief
Trace size for one run
32 MB. A larger trace isn’t saved, so Evals and Insights can’t read its steps; the run keeps its outcome.
What Evals read
,
What Insights read
Tasks, treatments, or agents in one experiment
100 each
Launch an experiment
Trials in one experiment
20
Launch an experiment
Runs in one experiment
100,000
Launch an experiment
Skills, MCP servers, CLIs, or SDKs in one treatment of a launch
100 each
Launch an experiment
Active schedules in one project
50
Schedules
Schedule start times
On the quarter hour: the cron minute is 0, 15, 30, or 45
Schedules
Schedule preview
The next 3 launches, and the runs of the next 30 days
Schedules
Operations in one
obss resources bulk-apply
20
obss resources bulk-apply
File or task import archive
512 MiB, 512 MiB expanded, and 65,535 files and folders
obss uploads create
Skill archive
256 MiB, 256 MiB expanded, and 65,535 files and folders
obss uploads create
Text file in a public experiment link
256 KiB. Larger and binary files aren’t shared.
Share results
Tag name
1 to 40 characters, with no comma
Tags
Tags on one record
20
Tags
Tags in one project
500
Tags
Records in one tag change
100
Tags
Tags named in one tag change or filter
20
obss tags
Custom agent name
120 characters
Custom agents
Models in one
set_models_enabled
or
set_model_context_windows
5,000
obss project ai apply
Connections in one
refresh_models
200
obss project ai apply
Runs in one
--requested-runs
check
1,000,000
obss quota inspect
Recent runs on a friction
10
Frictions
Next steps
CLI reference
Global flags, output, exit codes, and conventions.
Plugin (MCP) reference
IDs, paging, and polling from your AI assistant.
Glossary
Every Obss term in one place.
Run scope flags
Narrow runs and frictions by any record ID.
Tools
Powered by
This documentation is built and hosted on Mintlify, a developer documentation platform