v0.2.3

Changes

• Phonebook now exposes per-adapter model catalogs (id, call/personality routing notes, one default) and configured_model instead of a bare default_model string.
• Model descriptions guide TeLLMphone second-opinion routing (e.g. hostile review via Codex, architect via Claude opus/fable, cheap leaves for rubber-duck / tiny-hacker).
• Grok: install managed sandbox profiles and run headless turns with permission-mode auto so tools work under sandbox policy.
• Codex: pin TELLMPHONE_CALL_ID / hop into MCP config so progress reports work across resume.
• Install registers MCP with the active interpreter (python -m tellmphone.cli serve) for read-only sandbox friendliness.
• Accept legacy short call ids so one old record cannot break mailbox scanning.
• Voicemail answers return from / response consistently with live turns.

Upgrade notes

• Re-run tellmphone install after upgrading so Grok sandbox profiles and MCP serve commands are refreshed.
• Phonebook JSON shape changed: default_model → configured_model + models[]. Catalog defaults are suggestions only; they are not auto-applied when call omits model.

v0.2.2

• Allow headless Claude calls to use TeLLMphone tools under restrictive permission modes.

v0.2.1

• Hardened async call durability and access controls.
• Added progress updates and call transcript recovery.
• Improved self-calls, documentation, and tests.

v0.2.0

Add async filesystem-backed turns and Grok support.\n\n- Run live calls as detached one-shot tellmphone turn <call_id> processes instead of in-process worker threads.\n- Use turn.lock plus short call.lock critical sections so answers land durably in the mailbox even after the caller returns.\n- Preserve CLI wait behavior as polling over filesystem state.\n- Add Grok adapter and installer support.\n- Keep report_progress as a compatibility refusal stub while async mailbox delivery becomes the core long-task path.\n- Add coverage for detached turn startup, mailbox delivery, concurrent replies, failed turns, transcript locking, and Grok integration.\n\nVerified locally with uv run pytest and uv build.

v0.1.8

Fix Claude subprocess environment scrubbing so custom gateway/auth settings are preserved while host session markers are still removed.\n\n- Preserve Claude auth/provider/gateway environment variables such as CLAUDE_CODE_OAUTH_TOKEN, CLAUDE_CODE_USE_BEDROCK, ANTHROPIC_BASE_URL, and ANTHROPIC_AUTH_TOKEN.\n- Continue scrubbing parent Claude session markers that can misclassify child subprocesses.\n- Add regression coverage for the gateway environment case.

v0.1.7

What's Changed

• Add four built-in pragmatic engineering personalities:
  • the-algorithm
  • evidence-engineer
  • tiny-hacker
  • transaction-cost-accountant
• Update personality docs and tests.
• Bump package version to 0.1.7.

Verification

• uv run pytest

TeLLMphone v0.1.6

What's Changed

• Fix Gemini adapter prompt handling so agy receives the user prompt instead of --sandbox.
• Add the project directory to Gemini sessions so live calls can read repo files.
• Tighten Gemini adapter tests to match agy --print parsing.

TeLLMphone v0.1.5

Full Changelog: v0.1.4...v0.1.5

TeLLMphone v0.1.4

What's Changed

• Give every MCP tool a human-readable title so calling agents show a descriptive label (e.g. "Call another coding agent") instead of the bare tool name "call".
• Bump the mcp dependency floor to >=1.10 — the first release exposing the title kwarg on @mcp.tool() (absent in 1.9.4, present in 1.10.0).

Validation

• uv run pytest -q (66 passed)
• Verified all five tool titles propagate via list_tools()

TeLLMphone v0.1.3

What's Changed

• Add human CLI recovery commands: tellmphone messages, tellmphone show, and tellmphone gc.
• Add tellmphone call --mode voicemail and print call IDs for voicemail/ringing results.
• Allow same-agent calls so Codex/Claude can spawn a headless sibling session with a different personality or model.
• Keep top-level per-call write=True delegation while preventing callees from extending write grants down a call chain.
• Return full unread mailbox bodies from check_messages while retaining previews.
• Snapshot personality bodies in call records for transcript replay.
• Lengthen call IDs and validate mode / project_dir before creating calls.
• Add the neutral built-in personality.

Validation

• uv run pytest -q (66 passed)
• uv build