Hugging Face
Models
Datasets
Spaces
Buckets
new
Docs
Enterprise
Pricing
Website
Tasks
HuggingChat
Collections
Languages
Organizations
Community
Blog
Posts
Daily Papers
Learn
Discord
Forum
GitHub
Solutions
Team & Enterprise
Hugging Face PRO
Enterprise Support
Inference Providers
Inference Endpoints
Storage Buckets
Log In
Sign Up
Spaces:
smolagents
/
ml-intern
like
402
Running
on
CPU Upgrade
App
Files
Files
Community
46
Fetching metadata from the HF Docker repository...
fix-tool-errors-not-persisting
#2
by
elisaklunder
- opened
Apr 5
base:
refs/heads/main
←
from:
refs/pr/2
Discussion
Files changed
+8311
-42029
thinking if we want eval or not
2b6a5361
modified eval prompt
124a8a43
eval runs
4ce436f2
eval script done
af80aa78
removing jsonl csv
78e49c73
adding hf datasets i/o
3da9761b
adding readme
a1a22536
adding readme
f2e6e35b
minimal impl. of the agent
f730cddf
readme upt
08028f64
working events and session
8bff299b
demo cli
ef7b74ab
preliminary mcp
0d2b9f74
intermediary commit
0c252e46
mcp integration done, tools done, agent working
6b80d785
dataset creation script
f92b0c4e
readme
914a44af
readme
0127ea24
a working mcp integration
ecacd304
updated eval
035d1863
update deps
fe4d1289
link fix
1a8f5b2e
leaderboard and results
be350cbe
adding claude code + mcp
00d49dae
adding observability
e7068c0f
rename
7e214581
rename
2c820a40
fixing tracing
9de209dc
gpt 5 nano judge
9fe493be
readme update
9c4199bd
eval readme update
8f4b322a
intermediate commit until i let amp loose
158d846a
added compaction every x tokens
d574d65f
prompt change
fb5f9672
generated, filled in and verfied 250 eval questions
8541221a
gitignore and printing args
ea67d9f6
prompt update and loading from file
cd123ddf
custom hf jobs tool impl
4d8a7892
Fixed assistant -> assistant messages after compaction
86c3c8bf
Merge pull request #1 from huggingface/compacting-context
1675094f
ported the hf jobs tool to python
2ca459e0
adding the hf jobs tool to be exported
fc4f2473
lol yolo merge
af1a6640
Log streaming enabled
ff2fb511
small param / docs change to improve reliability of job scheduling by the agent
a82c70aa
pass ToolErrors to agent, correct HF token handling, simplify uv job calls
160da132
adding a user approval step to jobs api + ruff formatting
927e50ae
Merge pull request #3 from huggingface/job-tool-improvements
f29c8f5f
small prompt change, printing job url before stream instead of id
f84d68c5
added the explore subagent, quality ok but not perf
5e8489d7
Agent can now plan using a new tool
bab19bdb
re-order system instructions
c6444816
color for output, tool output truncation, plan printing
1297e911
prompt changes, more robust hf token sending, banner
916a4b84
added a api search tool and refined the hf docs search tools
eb923519
moved config file + small fixes
7291babf
Merge pull request #4 from huggingface/planning-and-design
83238f6f
Merge branch 'main' into explore-tool
1dfd3288
Merge pull request #5 from huggingface/explore-tool
65130640
formatting + typo fix
706d95dc
Resolve merge conflict in tools.py and update search agent config
75cb10f7
deleted help style call, added it to tool descritption and autoloaded HF token to env
99349180
improved search agent prompt and descriptions
a2e2d221
(partially done) system prompt tells to use research agent
90c3405c
reordered tools
b9c23252
reworked the hf-jobs tool
b63e2df9
reworked job tool descriptions + implemented batch processing for hf_job tool calls in cli
1598bb44
main agent search has better vibes
b307ff70
main agent search has better vibes
3765ba20
Added tools to create repos, view them, upload files to them (private). added a clock tool.
0c21cf1a
removed test
b6c2de68
updated jobs tool, system prompts and search capability
1766058a
made cli able to take in copy pasted things
a84363ca
Merge branch 'main' into private-repo-tool
b7b2c0c7
poc github tools
a6dcda8a
ruff
53d0a89e
github tools updated
63a4db3f
Merge pull request #8 from huggingface/private-repo-tool
46d481e5
find examples tool (repo crawling + fuzzy matching)
0b8bdf3f
list repos tool working
1d719d4c
unified ref tag and imporved read_file
caa90172
updated output for find examples
609f3e49
private tool handles spaces better + prompt improvements for reliability
c2cdd0d6
loading date in system prompt, search code update (still not there)
09727754
system prompt and github tool desc. update
bea39a38
improved search capabilities with github tools and system prompt updates
856b6192
removing extra prints
74736991
optional deps and main prints
7b4e2da0
readme update
d13cac75
small print removal
7fa8e8da
session logging
8d46a58e
subprocess uploading
f8d6755c
new system prompt and push to hub automatic checks
df460d9f
better file uploading
bc8323d5
comprehensive session logging
6a518b4b
Filter UV package installation output from HF Jobs logs
e5a148a8
jobs-tool-context-management
32f776ae
Add auto_file_upload and confirm_cpu_jobs config settings
4a6892ce
Deleted the example config file
68dc4d3e
enhanced prompt v0
5d182e4f
Added 2 dataset tools, one to list rows, the other one to download the dataset
ba44575c
cleaned comments
5ede083f
main merge
baee3791
modified timeout and rows parameters
56a36259
integrated hf skills into the prompts
0610df6d
Remove unused prompt files
4e3063a5
Merge pull request #10 from huggingface/prompt-and-reliability
078d4232
Show full job script in CLI instead of truncated snippet
e1198563
Add retry logic for job log streaming to handle connection drops
648dc32d
show-full-job-script
80319956
job-streaming-retry
f2578947
Add yolo mode for auto-approving tool calls
803b9667
fixing logs
67f8acee
jobs tool timeout errors;
e2b91f08
Unify dataset tools into single comprehensive hf_inspect_dataset tool
0d8b29e1
Prefer tool_calls or assistant message for example structure
33c04824
Fix example message selection and show as JSON with all keys
a8d5460e
Merge branch 'dataset_tools'
e96ab7e9
Merge configurable-permissions: add confirm_cpu_jobs, auto_file_upload, keep yolo mode
9bb2fc39
Simplify yolo_mode: use config instead of runtime variable
3b7a6b95
Fix: look for hardware_flavor at top level of tool_args, not nested
5d45d140
Merge pull request #14 from huggingface/dataset_tools
c37b0788
Merge remote main after PR #14
8688b668
Remove utils_tools and disable private_hf_repo_tool
c92b763a
Cleaning up tools
a2eb28d6
Remove skills/ from tracking
f9194c00
Merge disable-private-repo-tool
7efc07bd
Add hf_repo_files and hf_repo_git tools
8b2c9e34
Show file contents in hf_repo_files approval request
1dec1c0d
Add approval display details for hf_repo_files and hf_repo_git
5558a572
Added tool to go from draft to open PR, enabling merging + minor improvements
a2e414d8
Merge pull request #19 from huggingface/feat/hf-repo-tools
6692ff09
better data shape + small change in tool desc
6d8949db
limit in rows displaying
414e120a
small modif
6732b11b
Stop tracking session logs
a8a06cc0
adding whoami to system prompt
4197b96b
smol cleanup
91207892
whoosh impl for docs search
13025d53
adding token counts
9689843f
pre simple impl
914fa31f
file refactor, still works
9459bd97
Merge branch 'main' into dataset_tool_improved
471bd1ae
Merge pull request #20 from huggingface/dataset_tool_improved
e82b057a
ruff
6a4fd732
deleted search code
9d53405b
fix
0d853ecc
added whooooshing to api spec
e83f59a2
Add web frontend infrastructure for HF Agent
a56db97d
formatting
60a37550
Fix imports and add frontend to gitignore
d70b2088
Fix session_manager imports - define Operation/Submission locally
2fa71040
Fix config path to use absolute path from project root
0e6ce7ba
functioning frontend and docker
32f62c3e
format
1ae6efdc
adding frontend gitignore
f6432d87
added code panel, tool approval, tool calls logging
b1d24ded
hf jobs desc update
d3ffa60e
log streaming
226dea1d
improved design
29d492e0
plan tool bottom right
4485208b
logs have a button to see, improved rejection/approved flow, trying to handle tqdm
0bc8e384
improved tool calling, push to hub, etc.
aaf7d3bd
making docker run build work
fc14c05a
Improve job approval UX with tabs, job URL, and message ordering
b29799c1
Remove test files and build artifacts from git tracking
e74a635d
Merge branch 'main' into feature/web-frontend
a8796d4f
Merge pull request #21 from huggingface/feature/web-frontend
da441652
frontend update
bbfa4312
Track PNGs with LFS
78aac840
Fix directory name in README for cloning
4d92a9fa
Merge pull request #22 from abidlabs/patch-1
cb3b6caa
sandbox start
8d987c76
add create sandbox tool
82223926
sandbox working and agent can use it
8ded4a9a
enabled passing sandbox scripts to hf jobs
76996d00
adding compaction
0aec89b6
v3 prompt and tool desc rework
3f367da5
rewording
e4ae7cc6
Sandbox implementation
7b48ae0c
feat: merge HF Space improvements
bdbcdaba
feat: restore sandbox tools and proactive compaction
a33baefc
feat: use hf_agent tool descriptions, hardware flavors, and default env
c45ebae8
fix: use OAuth token for sandbox creation instead of env HF_TOKEN
1719ec54
feat: require sandbox_create approval before sandbox operations
eeab0b4a
fix: cast unknown to avoid TS2769 in sandbox approval UI
5969911b
feat: inject OAuth token as HF_TOKEN secret into sandbox Space
9ed57ca2
fix: remove TRANSFORMERS_VERBOSITY from sandbox env
9aba7d3b
feat: add hf_papers tool for paper discovery and reading
12c8f97c
fix: use correct HF API filter param and improve integration tests
c4344214
chore: temporarily disable MCP server
c9b25006
feat: implement cooperative agent cancellation with cancel button
2bc3b1a5
feat: cooperative cancellation, session persistence, background generation
3c77a6cd
feat: cooperative cancellation, session persistence, background generation
7e62dfac
feat: per-session WebSocket architecture for parallel chat sessions
854c2616
fix: approval UI not updating after approve click
f0c7e645
fix: reconnect WS on session switch instead of deleting dead sessions
a25dfc52
fix: route all HF token usage through session.hf_token from OAuth
5510397f
fix: use optional env var syntax for disabled MCP server config
61760c36
fix: inject OAuth token into MCP server headers at runtime
00102c43
fix: use user OAuth token for job creation (has job.write on org)
829ae5cd
feat: resolve sandbox file paths and show tool description in status
78f02010
feat: resolve sandbox file paths for hf_jobs script display
555a204f
fix: show activity status for approval-flow tools and revert chip override
9ba4cd40
fix: merge consecutive assistant messages on page refresh
56e52d83
fix: restore activity status bar when switching back to session
e0e8f9a3
feat: update models to MiniMax M2.5 (fireworks-ai) and Opus 4.6
8e0a54d3
feat: require org membership to create sessions
1d590c5e
feat: cleanup sandbox spaces on session end and increase cookie TTL
94e09153
feat: consecutive join/signin screens and sandbox status message
71196c04
fix: simplify onboarding to join-then-oauth flow
9ea884b6
fix: onboarding join org in iframe, sign-in on direct access
659c7463
fix: auto-detect org membership via localStorage, remove manual button
bdb3c28e
adding back mcp server
3b81a9cb
feat: input/output toggle in CodePanel, tool error handling
b1370a37
fix: input tab order and remove auto-scroll on panel open
93c72b5a
fix: buffer websocket events during disconnects and replay on reconnect
ea2014da
fix: increase sandbox sleep_time to 1500s
d36f7870
fix: improve compaction summary prompt, remove sidebar status bar
831b2caa
fix: remove unused StatusDot and isConnected from sidebar
1826db95
fix: only set sleep_time for non-CPU-basic sandboxes
c5c099cb
remove lmnr dependency
1de37c8c
replace websocket transport with SSE
571b292e
stream sandbox creation logs to tool output, fix hf_jobs final status race
da5e0c7a
fix: patch dangling tool calls on error, move undo to context manager
e9c82b71
fix: protect system prompt in undo, scan backwards for dangling tool calls, gate input on SDK status
f915e8e7
fix: show errorText for failed tools, bump eval max_iterations to 300
53a006d4
Revert "fix: show errorText for failed tools, bump eval max_iterations to 300"
dd8076b9
fix: properly close SDK message on error, show tool errorText
c68afb69
feat: upgrade CLI with local tools, slash commands, and interrupt support
392de349
fix: normalize dict tool_calls from litellm streaming to proper ToolCall objects
f7297087
feat: stream sandbox bash output to frontend in real-time
f97b6ec3
feat: show hardware pricing and sandbox explanation in tool approval UI
677b9d0b
fix: wire sandbox bash streaming to frontend panel
e35ce5d7
fix: reset approval UI state when new approval round arrives
cb3e47e0
fix: clear global activity state when switching away from a session
91130589
feat: add data audit section to system prompt
3ef54414
fix: recover malformed tool calls instead of crashing session
443a99ab
Revert "fix: wire sandbox bash streaming to frontend panel"
100eb8df
Revert "feat: stream sandbox bash output to frontend in real-time"
b94b18b2
fix: handle interrupt during streaming tool calls without corrupting context
a0178962
feat: improve bash tool status labels in activity bar
7ead77c5
fix: correct context_length init and emit tool_call events for malformed calls
085cd026
feat: prompt for HF token interactively when not found
7edb2252
refactor: simplify tool call validation and make interrupts cancel tool execution
960792d2
fix: gracefully handle MCP and OpenAPI failures during startup
04271776
fix: per-session state management for smooth task switching
723c24c3
fix: clear pending_approval before tool execution starts
e3e5ceb9
fix: disable chat input during processing after page refresh
c92e6dc6
feat: SSE keepalive, reconnection, and is_processing flag
00a57cd7
chore: increase sandbox sleep_time to 45 min
13c9794f
chore: double sandbox command timeouts
b3bfe9cd
fix: live event streaming after reconnection + LiteLLM timeout
c88804e9
fix: retry sandbox API calls on transient failures
28bdef89
fix: auto-retry on truncated tool calls + correct activity status on refresh
12b7c8fe
fix: use backend is_processing as source of truth on refresh
14163988
feat: kill sandbox processes and cancel HF jobs on user interrupt
9aea4e68
fix: prevent port conflict in HF Spaces dev mode
c99fbe6d
fix: exit gracefully on port conflict in dev mode
40c02f53
feat: cancel sandbox creation on user interrupt
0cee1987
feat: render cancelled tool state in UI after user interrupt
d2c1b128
Improve tool output quality: ANSI stripping, tail-biased truncation, noise suppression
6f67ddcd
feat: add research sub-agent tool, slim down main agent system prompt
1158f2cd
feat: emit tool_log events from research sub-agent for UI liveness
0b3071fb
feat: beautiful research sub-agent UI with inline progress steps
472f63cb
fix: update system_prompt_v3.yaml (the actual active prompt) to use research tool
a61953ee
fix: infinite re-render loop from Zustand selector returning new []
59b17bb8
fix: move import/signature validation to sandbox, keep host-side lightweight
7301cf53
Add deterministic pruning of old tool outputs before LLM compaction
06d26aaa
fix: skip research tool outputs in deterministic pruning
84633b2f
fix: also skip plan_tool from deterministic pruning
2383b4b6
feat: add doom-loop detection for repeated tool call patterns
0c2eb909
Add per-tool error limits and per-turn request cap
950ca5ce
refactor: append dynamic session context after static system prompt for cache stability
110c57a7
feat: LLM-based pruning for complex tool outputs (hf_jobs)
263e2296
Deduplicate file re-reads via content hashing for local and sandbox tools
0e5c34bc
feat: add headless CLI mode with local filesystem tools and rich terminal rendering
af6a7abd
Overhaul CLI visual design with rich-powered formatting
767d1028
Fix streaming: line-buffered output instead of per-token Live re-render
e97f3292
Replace compact panel with medium ASCII art banner
10417051
Fix Ctrl+C at idle prompt: no false interrupt, no stuck state
31086803
Tool output: last 10 lines for main agent, rolling 3-line display for sub-agents
2fc9917a
Hide research sub-agent tool_call and tool_output from CLI display
5b27d879
Switch banner to doom font with connected letterforms
3d9f1565
Buffer streamed chunks and render full markdown on completion
331c6834
Render streamed responses as markdown, indent agent output for alignment
55fc4e38
Indent all agent output, remove turn separator line
0a1e210c
Indent ASCII banner to match all other output
4eb48db4
Sub-agent display: 2 lines, gray, 20 char truncation
22c7ba4f
Log 20-char output preview for sub-agent tool calls
c9b20bbe
Sub-agent log: tool_name → output_preview on one line
f362c920
Revert "Sub-agent log: tool_name → output_preview on one line"
09860404
Show full sub-agent tool call lines, no truncation
2cc31b2c
Live research stats: tool count, tokens, elapsed time in header
5975e964
Research stats tick every second via background timer
fd8e4cda
Use Haiku for research sub-agent when main model is Anthropic
a80dea93
Use Haiku 4.5 for research sub-agent
267a9411
Only show plan_tool output in CLI, hide all other tool outputs
9f009a3c
Fix premature agent loop exit in long-running autonomous tasks
fdddeaaf
Merge pull request #31 from huggingface/agent-improvements
d95cff9f
Retry port binding in start.sh for dev mode restarts
9c6f4b2f
elisaklunder
Apr 5
No description provided.
elisaklunder
changed pull request status to
closed
Apr 5
Edit
Preview
Upload images, audio, and videos by dragging in the text input, pasting, or
clicking here
.
Tap or paste here to upload images
Comment
·
Sign up
or
log in
to comment