seektalent 0.3.4

Packaged v0.3.3 runtime with an interactive CLI agent for resume search

SeekTalent

English

SeekTalent is currently on the v0.3.3 active runtime. docs/v-0.3.3/ describes the active HEAD runtime and trace surfaces. docs/v-0.3.2/ is kept as the frozen pre-cutover baseline for comparison. HEAD ships deterministic requirement normalization, the bootstrap core, persistent-anchor frontier control, reviewer-ready output, and checked-in offline artifacts.

What exists today:

• docs/v-0.3.3/SYSTEM_MODEL.md is the active canonical spec for HEAD
• docs/v-0.3.3/IMPLEMENTATION_OWNERS.md is the active implementation anchor for HEAD
• docs/v-0.3.3/RUNTIME_SEQUENCE.md is the active timing view of the runtime
• docs/v-0.3.2/ remains frozen as the pre-cutover baseline
• src/seektalent/models.py holds the stable runtime contracts
• src/seektalent/requirements/normalization.py builds SearchInputTruth and normalized RequirementSheet
• src/seektalent/bootstrap.py runs the internal round-0 bootstrap flow
• src/seektalent/retrieval/filter_projection.py projects SearchExecutionPlan_t into CTS-safe native filters
• src/seektalent/clients/cts_client.py returns RetrievedCandidate_t
• src/seektalent/retrieval/candidate_projection.py builds SearchExecutionResult_t
• src/seektalent/runtime/orchestrator.py runs the full runtime loop and writes run artifacts
• seektalent run and run_match(...) return SearchRunBundle

What does not exist anymore:

• the old v0.2 controller / reflection / scoring / finalize runtime
• prompt bundles and LLM wiring from v0.2
• compatibility aliases for deleted contracts

Install

From a local checkout:

uv build
pipx install dist/seektalent-0.3.4-py3-none-any.whl

Or into an existing Python environment:

pip install dist/seektalent-0.3.4-py3-none-any.whl

Quick Start

Write a starter env file:

seektalent init

seektalent init writes the bundled starter template that ships with the package. The repo-root .env.example remains the authoring source for that template.

Minimal env values for real CTS mode:

OPENAI_API_KEY=your-openai-key
SEEKTALENT_CTS_TENANT_KEY=your-cts-tenant-key
SEEKTALENT_CTS_TENANT_SECRET=your-cts-tenant-secret

Each of the 5 LLM callpoints can now be switched independently through .env. See docs/configuration.md.

Run the local rerank API:

uv run --group rerank seektalent-rerank-api

Check the local runtime surface:

seektalent doctor
seektalent inspect --json

Human-first entry:

seektalent

This opens a chat-first Textual session in a TTY. Paste the JD, press Shift+Enter, then optionally add notes. The whole run streams into one transcript and the session ends after the final result.

Agent-friendly entry:

seektalent run --request-file ./request.json --json --progress jsonl

Minimal request file:

{
  "job_description": "Senior agent engineer with Python and LLM orchestration experience",
  "hiring_notes": "Shanghai preferred; startup background is a plus",
  "top_k": 10,
  "round_budget": 6
}

run --json still returns the full SearchRunBundle. The stable product result lives at final_result.final_candidate_cards.

Python API

The package exports run_match(...) and run_match_async(...) and returns SearchRunBundle:

from seektalent import AppSettings, run_match

result = run_match(
    job_description="Python agent engineer",
    hiring_notes="Shanghai preferred",
    settings=AppSettings(mock_cts=True),
    env_file=None,
)
print(result.final_result.stop_reason)
print(result.run_dir)

Commands

• seektalent (TTY only; launches the one-shot chat-first Textual session)
• seektalent run
• seektalent doctor
• seektalent init
• seektalent version
• seektalent update
• seektalent inspect

Docs

• docs/v-0.3.3/SYSTEM_MODEL.md
• docs/v-0.3.3/SYSTEM_MODEL.notion.md
• docs/v-0.3.3/IMPLEMENTATION_OWNERS.md
• docs/v-0.3.3/RUNTIME_SEQUENCE.md
• docs/v-0.3.2/SYSTEM_MODEL.md
• docs/v-0.3.2/IMPLEMENTATION_OWNERS.md
• docs/architecture.md
• docs/configuration.md
• docs/cli.md
• docs/outputs.md
• docs/development.md
• docs/_archive/v-0.3.1/implementation-checklist.md