laharikarumanchi/multi-tool-agent

Merge pull request #18 from laharikarumanchi-AI-ML/docs/spec-add-tool-use-failed

docs(spec): add tool_use_failed retry + _log_http_error_body sections (PR #17 follow-up)

Merge pull request #17 from laharikarumanchi-AI-ML/restore/groq-400-fix

Merge pull request #16 from laharikarumanchi-AI-ML/docs/spec-reality-audit

Merge pull request #15 from laharikarumanchi-AI-ML/feat/auto-dotenv

docs(llm): note tool_use_failed retry exception in chat_with_tools docstring

fix(llm): retry Groq HTTP 400 with code=tool_use_failed

fix(orchestrator): omit assistant `content` field when None alongside tool_calls

fix(llm): log response body on Groq HTTP error before raising

Merge pull request #14 from laharikarumanchi-AI-ML/chore/unused-imports-sweep

docs(spec): audit + reconcile spec vs built reality (6 drifts addressed)

test: stub load_project_env() in conftest to keep tests hermetic

feat(env): auto-load .env at CLI + eval entry points (no more `source .env`)

chore: remove unused imports across multitool/ and tests/

Merge pull request #13 from laharikarumanchi-AI-ML/revert-11-fix/groq-400-multi-turn

Revert "docs(llm): note tool_use_failed retry exception in chat_with_tools do…"

Merge pull request #12 from laharikarumanchi-AI-ML/revert-10-fix/groq-400-multi-turn

Revert "fix(llm): retry Groq HTTP 400 tool_use_failed (real root cause of 36% eval crashes)"

Merge pull request #11 from laharikarumanchi-AI-ML/fix/groq-400-multi-turn

docs(llm): note tool_use_failed retry exception in chat_with_tools docstring

Merge pull request #10 from laharikarumanchi-AI-ML/fix/groq-400-multi-turn

fix(llm): retry Groq HTTP 400 with code=tool_use_failed

fix(orchestrator): omit assistant `content` field when None alongside tool_calls

fix(llm): log response body on Groq HTTP error before raising

Merge pull request #9 from laharikarumanchi-AI-ML/feat/streamlit-demo

test(demo): smoke tests for app load, missing API key, example chips

chore(demo): HF Spaces frontmatter + requirements.txt note + .gitignore

feat(demo): Streamlit app for multi-tool agent

Merge pull request #8 from laharikarumanchi-AI-ML/feat/cli

fix(cli): treat --trace PATH as directory, always (avoid silent trace loss)

fix(cli): drop dead max-steps partial-answer branch + correct test contract

test(cli): cover argparse errors, missing key, stdout, verbose, max-steps

feat(cli): add multitool ask subcommand with argparse

Merge pull request #7 from laharikarumanchi-AI-ML/feat/eval

fix(eval): regex truncation, tolerance=None crash, empty-queries crash

feat(eval): run.py harness with per-task checkpointing

feat(eval): scorer with first-float-in-string heuristic

feat(eval): 25-query hand-curated test set, balanced across 5 categories

Merge pull request #6 from laharikarumanchi-AI-ML/feat/orchestrator

fix(agent): code-review polish — I-1, I-2 + 4 minor improvements

test(agent): end-to-end with scripted mock LLM

test(agent): per-dispatch retry budget + step-ceiling backstop

feat(agent): Orchestrator.run() — main function-calling loop

feat(agent): _dispatch_with_retry with 2-attempt budget

feat(agent): Orchestrator + AgentResult skeleton

feat(agent): Trace — JSON logger flushed after every step

feat(agent): system prompt for the orchestrator

Merge pull request #5 from laharikarumanchi-AI-ML/feat/llm-client

fix(llm): code-review polish — C1 + I1-I5 from PR #5 reviewer

feat(llm): GeminiClient.chat_with_tools() with synthesized call IDs