induscode 0.1.0__tar.gz

induscode-0.1.0/.gitignore

@@ -0,0 +1,21 @@
+# Python
+__pycache__/
+*.py[cod]
+*.egg-info/
+.eggs/
+build/
+dist/
+
+# Environments
+.venv/
+venv/
+
+# Tooling caches
+.pytest_cache/
+.ruff_cache/
+.mypy_cache/
+.coverage
+htmlcov/
+
+# OS
+.DS_Store

induscode-0.1.0/.pindusagi/settings.json

@@ -0,0 +1,3 @@
+{
+  "extensionPackages": []
+}

induscode-0.1.0/CHANGELOG.md

@@ -0,0 +1,72 @@
+# Changelog
+
+## 0.1.0 — 2026-06-11
+
+First release of the Python rebuild. Full port of the TypeScript
+`indusagi-coding-agent` (lineage **v0.1.62**, `indus-code-rebuild`) onto the
+Python `indusagi` framework, built milestone-by-milestone (M0–M6) per
+`indus-code-rebuild/PYTHON_PORT_PLAN/PLAN.md`. **649 tests green**, lineage
+scan clean, wheel verified in a fresh venv.
+
+### Subsystems
+
+- **workspace** — brand/version (single-sourced via `importlib.metadata`,
+  literal fallback) and the `~/.pindusagi` locator composed over the
+  framework `Locator` (`INDUSAGI_CODING_AGENT_DIR` > `INDUSAGI_HOME`).
+- **kit** — leaf utilities.
+- **settings** — two-tier `PreferenceStore` (project-write-only).
+- **briefing** — system-prompt composition, macro scanner, SKILL.md walker.
+- **window_budget** — context-window budgeting and condense seam
+  (3.6 chars/token heuristic kept).
+- **insight** — tracing wrapper over `indusagi.tracing` with collector sink,
+  replay, and FNV-1a gate.
+- **transcript_export** — SGR state machine, theme bridge, HTML publisher on
+  `markdown-it-py` + `pygments` (replacing `marked` + `highlight.js`).
+- **conductor** — turn loop with retry/queue-drain/persist-tail, condense
+  seam, fork/navigate, signal hub (10-tag translator table), model
+  catalog/matcher over `indusagi.ai`, branchable NDJSON `TranscriptStore`,
+  message⇄dict codec (`serialize.py`), skill parsing.
+- **runtime_bridge** — snapshot sink (mutable builder, per-push frozen
+  snapshots), broker (route/exchange/resume-tap), exchange driver + dialect
+  parsers.
+- **capability_deck** — builtin-bridge factory table, `PROFILE_TABLE`
+  provisioning, cards (todo ledger, bg-process daemon table on asyncio
+  subprocesses with SIGTERM→SIGKILL grace, task/saas/memory), bridge ledger
+  with ULID keys (`python-ulid`) and `mount_protocol_bridge` networking.
+- **addons** — discovery walk, importlib loader, event dispatcher
+  (gate-throw fails open, transform-throw keeps prior payload), host with
+  reserved names and first-wins conflicts.
+- **launch / boot / channels / sessions** — 17-flag invocation table on the
+  extended framework parser (`@file` attachments, `--`, list flags, `-pi`
+  clustering), credentials + OAuth adapter, multi-account `AuthVault` as the
+  single `auth.json` writer (0600, tolerant of framework single-record
+  shape), boot orchestrator + runners (oneshot/link/repl), NDJSON channel
+  framer (U+2028/29 escapes), session ops + link server + dialog bridge,
+  `SessionLibrary` over the transcript store.
+- **console / console_slash** — Textual TUI replacing the Ink console:
+  theme engine (4 schemes as native Textual themes), reducer, chrome
+  widgets, editor/keybindings with chord latch, autocomplete, overlays as
+  `push_screen` flows (incl. OAuth sign-in on `asyncio.Future`), and **26
+  slash commands** (18 transcript/workbench built-ins + 8 integration
+  commands, plus dynamic skill/template commands).
+- **entry** — console scripts `pindus` and `induscode`.
+
+### Key decisions
+
+- **Clean-room discipline** — independent implementation; no source copied
+  from any prior codebase; enforced by `scripts/lineage_scan.py` and the
+  public-API guard test, both release gates.
+- **Framework reuse over reinvention** — LLM abstraction, agent core, MCP
+  client, Textual shell primitives, and stage runner come from `indusagi`
+  (`[mcp,tui]` extras required); the agent keeps only its own seams
+  (transcript tree, vault, signal translation, deck).
+- **Behavioral parity with the TS lineage** — suites ported at full
+  strength (649 pytest cases vs the TS line's 332 vitest cases, never
+  weakened); versioning restarts at 0.1.0 while the TS line was at 0.1.62.
+- **Library swaps** — `marked`/`highlight.js` → `markdown-it-py`/`pygments`;
+  `ulid` → `python-ulid`; Ink/React → Textual; jiti virtual modules →
+  plain `importlib` addon loading.
+- **Packaging** — hatchling build; `py.typed` shipped; `NOTICE` and
+  `CREDITS.md` embedded via PEP 639 `license-files`; wheel gate = fresh
+  venv install (framework wheel first) + `--version`/`--help`/
+  `--list-models`/subsystem-import smokes.

induscode-0.1.0/CREDITS.md

@@ -0,0 +1,22 @@
+# Credits
+
+The **indusagi coding agent (Python rebuild, `induscode`)** is an independent implementation,
+designed and written from scratch against a behavioral specification and the public API of the
+`indusagi` framework — not by copying source from any prior codebase. It reuses no third-party
+application source code.
+
+## indusagi framework
+
+The agent runs entirely on the **indusagi** framework (Python rebuild, installed from
+`../indusagi-python-rebuild`), which provides the LLM API abstraction, agent core, Textual-based
+terminal-UI primitives, MCP client, and observability. The framework is itself an independent
+implementation and ships its own `CREDITS.md` / `NOTICE`.
+
+## Third-party runtime dependencies
+
+Standard open-source libraries used at runtime, each under its own license:
+`markdown-it-py` (transcript-export markdown rendering), `pygments` (transcript-export syntax
+highlighting), `python-ulid` (bridge-ledger keys), plus the framework's `[mcp,tui]` extras
+(`mcp`, `textual`). Dev/test: `pytest`, `pytest-asyncio`, `build`.
+
+See `NOTICE` for the notice that must travel with redistributions.

induscode-0.1.0/GAP_REPORT.md

@@ -0,0 +1,150 @@
+# induscode Python Port — Gap Report
+
+**Date:** 2026-06-12
+**Ground truth (TS):** `/Users/varunisrani/indusagi-ts/indus-code-rebuild`
+**Port under audit (Python):** `/Users/varunisrani/indusagi-ts/induscode-python-rebuild`
+
+## Executive verdict
+
+The Python port is at **near-full functional parity** with the TS ground truth. All 15 subsystems are present (none absent), the entire user-facing surface — 26/26 slash commands, every CLI flag, all console overlays, settings, sessions — matches, and 649 Python tests pass against the same `PYTHON_PORT_PLAN/PLAN.md` baseline the TS rebuild ships. After applying the critic's corrections (one severity downgrade, zero gaps disproven), **7 real gaps remain: 0 high, 1 medium, 6 low** — an estimated **~98% behavioral parity**. The gaps cluster entirely in two subsystems: `launch` (CLI surface) and `insight` (observability/tracing). Nothing in the core interactive agent loop, tool execution, or console UI is broken.
+
+**The single most important thing to fix:** the CLI `--resume` flag's interactive session picker. On a real TTY, `pindus --resume` can never present the session list — `_DefaultResumeDeps.mount_picker` raises and the runner silently falls back to a fresh session. It is *medium* (not high) only because the in-console `/resume` slash command is fully wired and reaches the exact same arbitrary-session picker, so the capability is not lost — just degraded at one entry point.
+
+---
+
+## Gap summary table
+
+Sorted severity high → low. The critic disproved no gaps; the only correction was downgrading gap #1 from high to medium (capability is reachable via `/resume`).
+
+| Feature | Subsystem | Status | Severity |
+|---|---|---|---|
+| CLI `--resume` interactive session picker falls back to fresh | launch | broken | **medium** |
+| GitHub Copilot browser sign-in provider | launch | missing | low |
+| Insight secret redaction over-redacts (whole value vs matched run) | insight | divergent | low |
+| Insight `fail()` does not emit an in-flight update signal | insight | divergent | low |
+| Insight persisted/streamed fault loses error name + stack | insight | partial | low |
+| Insight redactor `omitKeys` option (drop keys wholesale) | insight | missing | low |
+| Top-level package namespace re-export of all subsystems | root barrel | divergent | low |
+
+> **Note — not a port gap (DO-NOT-FIX):** the `-p`/`--json`/NDJSON first-chunk drop ("pong"→"ong") is an **upstream framework** defect in `indusagi-python-rebuild/.../agent/agent.py` (`_FacadeTranslator._on_delta` emits the first delta only inside the `MessageStartEvent` partial). The app side (`conductor/signal_hub.py`, `message_start → []`) is parity-correct with TS `translate.ts`, and the TUI masks it via the final `message_end`. Filed upstream; excluded from this table.
+
+---
+
+## Per-subsystem gaps
+
+### launch (CLI surface)
+
+#### 1. CLI `--resume` interactive picker never mounts on a TTY — *medium, broken*
+
+**TS evidence:** `indus-code-rebuild/src/launch/pickers.ts:56-99` — `defaultResumeDeps().mountPicker` renders the real React-Ink `StartupSessionPicker`; the repl runner calls `pickResumeTarget(load, load)` with no overrides and the live picker mounts on a TTY.
+
+**Python evidence:**
+- `induscode-python-rebuild/src/induscode/launch/pickers.py:95-105` — `_DefaultResumeDeps.mount_picker` unconditionally `raise RuntimeError("The interactive session picker is wired by the console runner.")`.
+- `…/boot/runners/repl_runner.py:223` — `_choose_resume_target` calls `pick_resume_target(load, load)` with **no injected deps**; grep confirms the raising default is the only `mount_picker` anywhere.
+- At runtime the interactive path returns a typed fault and `repl_runner.py:270` prints `"Resume failed ({detail}); starting a fresh session."` (the framework's Textual `StartupSessionPicker` exists but is unused on this path).
+
+**User impact:** On a real terminal, `pindus --resume` always silently starts a brand-new session (one-line stderr notice). The user cannot pick an arbitrary prior session from the CLI flag. **Mitigated:** the in-console `/resume` command is live — `console/slash_commands/transcript.py:196` (`ctx.open_modal("sessions")`) → `console/overlays/router.py:96` (`"sessions": run_session_picker`) → `console/overlays/sessions.py` uses `app.push_screen_wait(...)` on a real Textual session dialog. So a user can run `pindus --resume`, land in a fresh session, type `/resume`, and get the exact arbitrary-session picker. Capability degraded at one entry point, not lost.
+
+**Fix hint:** In `repl_runner._choose_resume_target`, inject a Textual-backed `ResumeDeps` whose `mount_picker` reuses the already-wired `run_session_picker` (push it as a modal screen, resolve from its select/close), then pass that as the deps arg to `pick_resume_target` instead of relying on the raising default. The fix is small — the picker plumbing already exists for `/resume`.
+
+#### 2. GitHub Copilot browser sign-in provider missing — *low, missing*
+
+**TS evidence:** `indus-code-rebuild/src/launch/oauth.ts:47-50` — `BUILT_IN_OAUTH` registers three framework providers: `anthropicOAuthProvider`, `openaiCodexOAuthProvider`, `githubCopilotOAuthProvider`.
+
+**Python evidence:** `…/launch/oauth.py:197-213` — `register_built_in_oauth_providers` walks the framework `OAUTH_PROVIDERS` table, which contains only `['anthropic', 'openai']`; `list_login_providers()` emits 2 OAuth rows. The framework `indusagi.llmgateway.credentials.oauth.OAUTH_PROVIDERS` has no `github-copilot` config (Copilot appears only in model-catalog data). The app-side label is already present: `oauth.py:169` `"github-copilot": "GitHub Copilot"`.
+
+**User impact:** `pindus signin github-copilot` resolves to an unknown provider; the merged sign-in directory shows 2 of TS's 3 rows. No browser sign-in for Copilot.
+
+**Fix hint:** Add a `github-copilot` `OAuthConfig` to the framework `indusagi.llmgateway.credentials.oauth.OAUTH_PROVIDERS` table. Because `_OAUTH_LABELS` already carries the label, `register_built_in_oauth_providers` picks it up automatically once the config exists. This is a framework-data gap, not app logic.
+
+### insight (observability / tracing)
+
+> Context: per waiver **W4**, the bulk tracing engine stays in the Python framework (`indusagi.tracing`); only the collector sink, replay reader, and the redaction/recorder wrapper are app code. All four insight gaps below are wrapper-side costs of delegating to the framework `SecretScrubber` / `SegmentError` / transport, and are documented inline in `insight/wrapper.py`. None affects the final close record's status or any end-user agent behavior; impact is limited to live trace dashboards and offline trace analysis.
+
+#### 3. Secret redaction over-redacts: whole value instead of matched run — *low, divergent*
+
+**TS evidence:** `indus-code-rebuild/src/insight/redaction.ts:124-137` — `applyValueRules` does `out.replace(pattern, REDACTED_TOKEN)`, replacing only the matched substring; `insight.test.ts:270-272` asserts `scrubValue("Authorization is Bearer abcdef…")` still contains the `"Authorization is "` prefix.
+
+**Python evidence:** `…/insight/wrapper.py:481-484` — `scrub_value` delegates to the framework `SecretScrubber`, which replaces the entire value. At runtime `scrub_value("Authorization is Bearer abcdef…")` returns `"[insight:scrubbed]"` (whole string).
+
+**User impact:** Trace attributes that embed a secret inside prose lose all surrounding non-secret context in persisted/streamed traces. Over-redacts (safe) but degrades trace readability vs TS.
+
+**Fix hint:** Re-port the TS `applyValueRules` (per-rule `re.sub`, matched-run only) wrapper-side in `SecretRedactor.scrub_value` instead of delegating wholesale, or extend the framework `SecretScrubber` with a matched-run mode. (Documented as W4 cost at `wrapper.py:56-58`.)
+
+#### 4. `fail()` emits no in-flight update signal — *low, divergent*
+
+**TS evidence:** `indus-code-rebuild/src/insight/recorder.ts:216-224` — `fail()` sets `status='error'` + fault, then `route({phase:'update', …})` immediately, so a live sink observes the failure before close.
+
+**Python evidence:** `…/insight/wrapper.py:669-674` — `fail()` only stashes `self._fault` and emits nothing (the framework transport has no probe-carrying update signal). At runtime a failing child probe produces no `update` between its `open` and `close`.
+
+**User impact:** A live `signals()`/stream-sink observer (e.g. a real-time trace dashboard) sees a probe's failure only when it closes, not the moment `fail()` is called. No effect on the final close record.
+
+**Fix hint:** Have `_LiveProbeHandle.fail()` emit a synthetic `update` through the channel (reconstructing the error-status probe), matching `recorder.ts`. (Documented as W4 cost at `wrapper.py:50-54`.)
+
+#### 5. Persisted/streamed fault loses error name + stack — *low, partial*
+
+**TS evidence:** `indus-code-rebuild/src/insight/recorder.ts:314-326` — `faultOf` captures `message + name + stack`; `serialize.ts:64-72` persists the full probe incl. `fault.name/stack`; `redaction.ts:223-234` `scrubFault` preserves name and scrubs+caps stack.
+
+**Python evidence:** `…/insight/wrapper.py:208-219` — `fault_of` captures name+stack on the in-memory handle, but `close()` (`wrapper.py:683-687`) builds the framework `SegmentError(message=…)` only; `replay.py:174-177` `record_to_probe` rebuilds `fault=ProbeFault(message=…)` with `name=None, stack=None`. At runtime the CloseSignal fault is `ProbeFault(message='boom', name=None, stack=None)`.
+
+**User impact:** Persisted NDJSON traces (and the in-process close signal) lose the error type name and stack trace. Offline trace analysis in Python sees only the message; TS retains exception class + stack.
+
+**Fix hint:** Either widen the framework `SegmentError`/replay schema to carry `name`/`stack`, or stash them in the close-record attributes wrapper-side and recover them in `record_to_probe`. (Documented as W4 cost at `wrapper.py:50-54` / `replay.py`.)
+
+#### 6. Redactor `omitKeys` option missing — *low, missing*
+
+**TS evidence:** `indus-code-rebuild/src/insight/contract.ts:470-474` defines `RedactionOptions.omitKeys`; `redaction.ts:155-220` `createRedactor` honors it — matching keys are deleted from the scrubbed output entirely (not tokenized).
+
+**Python evidence:** `…/insight/wrapper.py:538-544` — `create_redactor` takes only `patterns` + `max_string_length`; the framework `SecretScrubber.__init__` signature is `(patterns, token)` with no omit support (confirmed via `inspect`); `SecretRedactor` has no omit path.
+
+**User impact:** Callers cannot configure keys to be dropped from trace output (only tokenized). In practice TS never wires actual `omitKeys` in production code, so the default redactor behaves identically — pure API-surface gap.
+
+**Fix hint:** Add an `omit_keys` parameter to `create_redactor`/`SecretRedactor` and delete matching keys before/after the `SecretScrubber` pass (the scrubber walks but never removes keys). Low priority — unused in TS production paths.
+
+### root barrel (public library surface)
+
+#### 7. Top-level namespace re-export of all subsystems missing — *low, divergent*
+
+**TS evidence:** `indus-code-rebuild/src/index.ts:9-25` re-exports all 15 subsystems as namespaces (`export * as runtimeBridge from "./runtime-bridge";`, and likewise for every other subsystem).
+
+**Python evidence:** `induscode-python-rebuild/src/induscode/__init__.py:8-18` re-exports only `workspace` (+ `VERSION`/`__version__`); `__all__` lists just those. The barrel docstring still references the M0-era stub state.
+
+**User impact:** A consumer cannot do `import induscode; induscode.runtime_bridge…` the way TS allows `agent.runtimeBridge…`. The idiomatic Python path `from induscode.runtime_bridge import …` works identically, so there is no behavioral consequence. This is a build-wide barrel convention difference, not specific to any one subsystem.
+
+**Fix hint:** If top-level namespace access is desired, add `from induscode import <subsystem>` imports plus the names to `induscode/__init__.py` `__all__`. This is a cross-cutting decision affecting all 15 subsystems; refresh the stale M0 barrel docstring at the same time. Low priority — no functional gap.
+
+---
+
+## Deliberate / accepted differences (waived ports — intentional, not bugs)
+
+These 13 differences are plan-cited waivers in PARITY_REPORT. They are by-design ports of TS behavior onto Python/Textual/framework idioms, fully tested, and **not** defects:
+
+- **W1 — Composer deleted.** The TS bespoke composer is replaced by the framework `PromptEditor`; folded into `console/app.py` + `console/input/`. Structural, not missing.
+- **W2 — Paste vault / burst-debounce deleted.** Replaced by framework `EditorCore` paste markers.
+- **W3 — jiti → importlib.** `.py` addons load via `spec_from_file_location` (`addons/loader.py`); the TS virtual-module/alias bridge collapses and `BUNDLED_NAMESPACES` is vestigial.
+- **W4 — insight wraps `indusagi.tracing`.** ~2.5k LOC of tracing stays in the framework; only collector sink + replay reader + FNV-1a gate are app code. (Explains the lower Python insight LOC and gaps #3–#6 above.)
+- **W5 — repl placeholder → real Textual mount.** The M0 `cli.py` stub is replaced by a real console mount in `boot/runners/repl_runner.py` (the seam stays injectable).
+- **W6 — runtime-bridge barrel-only.** A tested library layer in **both** builds; deliberately not wired into the boot path in either.
+- **W7 — `/memory` on/off cosmetic.** Flips a reporting flag only; never removes the card. Matches TS.
+- **W8 — two distinct compaction engines.** `window_budget` (tokens, 0.75 threshold, 3.6 chars/token) is the wired conductor `CondenseFn`; the framework `runtime.memory` (turns, 0.8) is intentionally unwired — same as TS.
+- **W9 — chord-latch break semantics.** Timer-bounded under Textual; behavioral parity with the TS chord latch.
+- **W10 — OAuth overlay on `asyncio.Future`.** The OAuth overlay resolves via a Future instead of the TS promise; same UX.
+- **W11 — TypeBox → literal JSON-schema dicts.** Addon/tool schemas are plain dicts instead of TypeBox builders.
+- **W12 — modal routing inversion.** TS always-mounted overlays become `push_screen` via `console/overlays/router.py`'s `OVERLAY_HANDLERS` table.
+- **W13 — todo wire names.** `todo_read` / `todo_set` (vs TS internal names); functionally identical.
+
+**Out of scope (TS feature ceiling, not a Python parity baseline):** `FEATURE_GAP_ROADMAP.md`, `UI_IMPROVEMENTS_FROM_INDUSCODE.md`, `UI_UPGRADE_PLAN.md`, `CHANGELOG.md` describe the gap between the TS rebuild and the upstream `induscode` Claude-Code fork (markdown→ANSI rendering, colored diffs, syntax highlight, read-before-edit/mtime gate, deep-recursive grep, broken-`find`-glob fix, LSP diagnostics, plan mode, CLAUDE.md ingestion). Both rebuilds wire the **same** framework `facade/bot/actions` built-ins (verified: Python `capability_deck/builtin_bridge.py` + `manifest.py` wire identical read/write/edit/ls/grep/find built-ins), so these affect TS and Python equally and are **not** Python parity gaps.
+
+---
+
+## Recommended fix order
+
+Highest user-impact first.
+
+1. **Wire the CLI `--resume` Textual picker** (gap #1, medium). In `boot/runners/repl_runner.py:_choose_resume_target`, inject a `ResumeDeps` whose `mount_picker` reuses the already-live `run_session_picker` (push a modal Textual screen, resolve from select/close) and pass it to `pick_resume_target(...)`. Removes the silent fresh-session fallback on `pindus --resume`. Small fix — the picker already exists for `/resume`.
+2. **Add the GitHub Copilot OAuth config** (gap #2, low). Add a `github-copilot` `OAuthConfig` to the framework `indusagi.llmgateway.credentials.oauth.OAUTH_PROVIDERS`. App label is pre-wired, so registration is automatic. Restores browser sign-in for Copilot.
+3. **Port matched-run secret redaction** (gap #3, low). Re-implement `applyValueRules` (per-rule `re.sub`, matched run only) in `SecretRedactor.scrub_value` instead of delegating wholesale to `SecretScrubber`. Restores trace readability around embedded secrets.
+4. **Persist fault name + stack** (gap #5, low). Stash `name`/`stack` in the close-record attributes wrapper-side and recover them in `replay.record_to_probe` (or widen the framework `SegmentError`/replay schema). Restores offline-debug fidelity.
+5. **Emit in-flight `fail()` update** (gap #4, low). Have `_LiveProbeHandle.fail()` push a synthetic `update` signal so live trace dashboards see failures before close.
+6. **Add `omit_keys` to the redactor** (gap #6, low). Add the parameter to `create_redactor`/`SecretRedactor` and delete matching keys around the scrubber pass. Closes the API-surface gap (unused in TS production).
+7. **Restore the top-level subsystem barrel** (gap #7, low). Re-export all 15 subsystems from `induscode/__init__.py` and refresh the stale M0 docstring. Cosmetic library-surface parity; no behavioral change.

induscode-0.1.0/NOTICE

@@ -0,0 +1,7 @@
+induscode — indusagi coding agent (Python rebuild)
+Copyright (c) 2026 indusagi
+
+This product is an independent implementation, designed and written from scratch.
+It reuses no third-party application source code.
+
+This agent runs on the indusagi framework, which ships its own NOTICE.

induscode-0.1.0/PARITY_REPORT.md

@@ -0,0 +1,173 @@
+# PARITY_REPORT.md — induscode Python rebuild vs TS ground truth
+
+**Date:** 2026-06-11 · **Milestone:** M6 (parity-audit half)
+**Python build:** `/Users/varunisrani/indusagi-ts/induscode-python-rebuild`
+**TS ground truth:** `/Users/varunisrani/indusagi-ts/indus-code-rebuild`
+**Checklist audited:** `indus-code-rebuild/PYTHON_PORT_PLAN/PLAN.md` §6, row by row, against actual code + tests (not file names).
+
+## Summary
+
+| Metric | Result |
+|---|---|
+| **Verdict** | **PASS — full feature parity.** 106/106 checklist rows **done**; 0 partial, 0 missing. 13 deliberate, plan-cited waivers/divergences documented below. 2 plan-doc errata (not port gaps). |
+| Test suite | **649 passed**, 0 failed (`.venv/bin/python -m pytest -q`, 8.07 s) |
+| Fresh-venv wheel gate | **PASS** (packaging half of M6, per the packaging agent) |
+| Live smoke (a–d) | 4/4 executed: (b) (c) (d) clean PASS; (a) exit 0 with one **upstream framework defect** — first streamed chunk dropped from `-p` text output (root cause below; not fixed, per instructions) |
+| Rows by area | flags 17 + conventions 6 · slash 26 static + 4 dynamic + 10 aliases · overlays 12 + console-other 4 · channels 5 · addons 4 + deck 4 · sessions 8 · runtime-bridge 2 · aux 4 = **106** |
+
+---
+
+## Live smoke results (owner's key; `INDUSAGI_HOME=/tmp/induscode-live-test`)
+
+| # | Command | Expected | Result |
+|---|---|---|---|
+| a | `.venv/bin/pindus -m claude-haiku-4-5 -p "reply with the word pong"` | `pong`, exit 0 | **PASS w/ DEFECT** — exit 0, model resolved (`claude-haiku-4-5` accepted by catalog/matcher; no resolution issue), but stdout was `ong\n`. Reproduced twice; a second probe (`reply with exactly: strawberry banana`) printed `berry banana` — the **entire first streamed chunk** is dropped, not a fixed character count. Root cause below. |
+| b | `echo '{"jsonrpc":"2.0","id":1,"method":"snapshot"}' \| .venv/bin/pindus --json` (12 s harness, kill after reply) | one JSON-RPC reply line | **PASS** — `{"jsonrpc":"2.0","id":1,"result":{"model":…,"sessionId":"s_…","usage":{…}}}` — a flat `LinkSnapshot`; process exited 0 on stdin close. |
+| c | TUI under a pty (`pty.fork`, 120×40, wait ~5 s, prompt, wait ~8 s, `/quit\r`) | banner wordmark, streamed text, clean exit | **PASS** — `INDUS CODE` block-art wordmark + `indus console v0.1.0` rendered; session card showed `model claude-haiku-4-5`; streamed reply arrived (transient `ong` during streaming, **final render `pong`** — the defect is masked in the TUI because `message_end` carries the full message); `/quit` exited with status 0. Capture: 57,009 bytes. |
+| d | `pindus signin --list` (sandboxed) | exit 0 | **PASS** — printed `Saved accounts:\n  (none)`, exit 0. |
+
+Side evidence from the live run: sessions were written to `/tmp/induscode-live-test/sessions/--Users-varunisrani-indusagi-ts-induscode-python-rebuild--/s_*.ndjson` with first line `{"schema":"indus/transcript@1",…}` — live confirmation of per-cwd `--<slug>--` scoping, the transcript header, and `INDUSAGI_HOME` honoring.
+
+### Root cause: dropped first streamed chunk (smoke a) — DO-NOT-FIX filed
+
+- **Where:** framework, `/Users/varunisrani/indusagi-ts/indusagi-python-rebuild/src/indusagi/agent/agent.py`, `_FacadeTranslator._on_delta` (line 915).
+- **Mechanism:** on the **first** text delta of an assistant message, `_message_open` is false, so the translator emits only `MessageStartEvent(message=partial)` (line 924) — the first chunk's text travels **only inside the partial message**. `MessageUpdateEvent(assistantMessageEvent={"type":"text_delta","delta":…})` is emitted for second-and-later deltas only (lines 926–931, the `else` branch).
+- **App side (parity-correct):** induscode's translator maps `message_start → []` (`/Users/varunisrani/indusagi-ts/induscode-python-rebuild/src/induscode/conductor/signal_hub.py` line 355 — same as TS `translate.ts`), so no `TextSignal` is ever produced for the first chunk. The oneshot text strategy (`channels/oneshot.py`, `_text_strategy`) accumulates only `TextSignal` deltas → `-p` output is missing the first chunk ("pong" → "ong").
+- **Blast radius:** `-p` text output (truncated), `-p` NDJSON signal stream and `--json` link signal stream (first text chunk absent from the `text` signal sequence), TUI in-flight stream (transient; final render is correct because `message_end` carries the complete assistant message).
+- **Proposed fix (not applied):** in `_FacadeTranslator._on_delta`, when opening the message emit `MessageStartEvent` with an **empty** partial and then *always* emit the `MessageUpdateEvent` carrying the delta (i.e., move the update emit out of the `else`). That matches the TS framework's framing, where the first delta arrives as a `message_update`. An app-side workaround (extracting trailing text from the `message_start` partial in `signal_hub.py`) would diverge from the TS translator table and is not recommended.
+
+---
+
+## `--tools` `_`/`-`-insensitivity: settled verdict
+
+**DONE — full parity; the plan row's wording is misattributed, not wrong.** Evidence:
+
+- **TS:** `src/launch/invocation/read.ts` `resolveTools` (line 252) filters the parsed list to **exact** roster names via `isToolName` — no normalization at parse time. The normalization lives at the deck-filter seam: `src/boot/runners/session.ts` line 261, `const canon = (name) => name.toLowerCase().replace(/[_-]/g, "")` inside `selectTools` — it exists so a `--tools web_fetch,todo_read` request lines up with the deck's `webfetch`/`todoread` ids (comment at lines 252–254).
+- **Python:** identical at both seams. Parse-time exact filter: `launch/invocation/read.py` `_resolve_tools` (line 265) + `launch/contract.py` `is_tool_name` (line 198) over the identical 13-name `TOOL_NAMES` roster (contract.py:181 ↔ contract.ts:147). Deck-filter canon: `boot/runners/session.py` `_canon_tool_name` (line 314, `re.sub(r"[_-]", "", name.lower())`) used in `select_tools` (lines 320–330) — including the same quirk that an allow-list emptied by parse-time filtering falls back to **all** tools (session.py:327 ↔ session.ts:260).
+- **Plan-doc erratum:** PLAN.md §6 attributes the insensitivity to `launch/invocation/flags.py`; in **both** codebases it lives in the boot session runner's deck filter. The CLI auditor's PARTIAL flag is overturned.
+
+---
+
+## Audit tables (condensed; full evidence cited per row)
+
+### 1. CLI flags (17) + conventions (6) — 23/23 done
+
+| Row | Status | Evidence (Python · test) |
+|---|---|---|
+| `--print`/`-p` oneshot | done | `launch/invocation/flags.py:96` ; `_derive_mode` `read.py:243-253` · `tests/launch/test_launch.py:187-218` |
+| `--json`/`--rpc` wins over `-p` | done | `flags.py:103`; json checked before print in `_derive_mode` · `test_launch.py:220-223`, `tests/boot/test_invocation.py:71-75`; live smoke (b) |
+| `--interactive`/`-i` force REPL | done | `flags.py:110` · `test_launch.py:226-229` |
+| `--model`/`-m` | done | `flags.py:118` · `tests/boot/test_boot.py:62-66`; live smoke (a/c) |
+| `--account` | done | `flags.py:125` · `test_boot.py:20-25` |
+| `--thinking` 6-level | done | `flags.py:131` + `THINKING_EFFORTS` (off/minimal/low/medium/high/xhigh) `contract.py:139-146` · `test_boot.py:20-25` |
+| `--list-models [substr]` | done | `flags.py:137`; filter projection `boot/invocation.py` · `test_invocation.py:49-57` |
+| `--cwd` | done | `flags.py:144` · `test_boot.py:83-87` |
+| `--resume`/`-r` picker | done | `flags.py:162`; `boot/invocation.py:87` · `test_invocation.py:42-47` |
+| `--continue`/`-c` newest-in-cwd | done | `flags.py:169`; `boot/invocation.py:88` · `test_invocation.py:42-47` |
+| `--tools` (13-name roster, `_`/`-`-insensitive) | **done** | settled above: `contract.py:181-200`, `read.py:265-271`, `boot/runners/session.py:314-330` · `test_boot.py:33-36` |
+| `--no-tools` | done | `flags.py:183`; `read.py:345` · `test_boot.py:33-36`, `test_launch.py:200-201` |
+| `--mcp` config-path list | done | `flags.py:189`; accumulate + comma-split `read.py:126-175` · `test_launch.py:193-196`, `test_invocation.py:38-40` |
+| `--help`/`-h` from the table | done | `flags.py:196`; `render_usage` walks `FLAG_SPECS` `usage.py:72-110` · `test_boot.py:99-105`, `test_launch.py:237-248` (FLAG_SPECS↔usage coverage) |
+| `--version`/`-v` → `induscode 0.1.0` | done | `flags.py:203`; `boot.py:91-94` `f"{BRAND.name} {VERSION}"`; `workspace/brand.py:66-96` · `test_boot.py:107-112` |
+| `--system` file-or-literal | done | `flags.py:150`; `resolve_text` in session runner · `test_boot.py:27-31` |
+| `--append-system` file-or-literal | done | `flags.py:156` (hyphenated bag key) · `test_boot.py:27-31` |
+| `@file` attachments (text/image, caps, typed errors) | done | `invocation/attachments.py:42-150` (10 MB text / 20 MB image; not-found/too-large/unsupported/read-failed) · `test_launch.py:256-283` |
+| `--` terminator | done | `read.py:300-303` + `read_file_references` re-walk · parser tests |
+| `-pi` clustering | done | `_is_short_cluster`/`_apply_short_cluster` `read.py:71-74, 210-222` (booleans only) · integration parses |
+| unknown-flag tolerance → flag bag | done | `read.py:327-330` extension-flag hatch |
+| `signin\|login`/`signout\|logout` + `--provider/--account/--method/--oauth/--api-key/--list` | done | `launch/credentials.py:343-366`; OAuth adapter `launch/oauth.py` · `test_launch.py:343-381`; live smoke (d) |
+| `install/remove/update/list/config` packages | done | `launch/packages.py:49-57` over framework parser engine · `tests/launch/test_packages.py` |
+
+### 2. Slash commands — 26 static + dynamic + aliases, all done
+
+| Group | Status | Evidence |
+|---|---|---|
+| Transcript ×11 (`/clear /new /summarize-context /resume /session /branch /timeline /name /reload /quit /exit`) | done | `console/slash_commands/transcript.py:343-403` (handlers L133-331; tracked fire-and-forget `_spawn` L87-91) · `tests/console/test_commands_transcript.py` (14 cases) |
+| Workbench ×7 (`/model /models-for /settings /help /keys /whats-new /debug`) | done | `workbench.py:387-430` (21-row HOTKEYS table; late-bound help registry; /debug JSONL via message⇄dict codec) · `test_commands_workbench.py` (13 cases) |
+| Integrations ×8 (`/login /logout /mcp /memory /composio /copy /export /share`) | done | `integrations.py:901-949`; per-console `IntegrationsRuntime` holder L102-138; `/share` runs `gh gist create` and degrades to local HTML path on missing/unauth gh · `test_commands_integrations.py` (18 cases) |
+| Dynamic `/skill:<name>` + `/<template>` macros + collision filtering | done | `dynamic.py:129-176` (skill XML block w/ attr escaping; `$1/$2/$ARGUMENTS` + `{{arg.*}}` macro expansion via briefing/macros); self-collision dedup `dynamic.py:202-209`; reserved-token filter `builtins.py:196-206` · `test_commands_dynamic.py` |
+| Aliases ×10 (`reset · condense/compact · sessions · fork · tree · models · scoped-models · ? · hotkeys · changelog`) | done | alias tuples on each `SlashCommand` (transcript.py:346,357,364,375,381; workbench.py:392,398,411,417,423) · alias-coverage test in each group |
+
+### 3. Overlays, theme, input, startup — 16/16 done
+
+| Row | Status | Evidence |
+|---|---|---|
+| 12 ModalKinds: models · scopedModels (edit/show/reset) · settings · theme · sessions · tree · userTurns · signIn (provider threading) · signOut · oauth (full flow on `asyncio.Future`) · plugin (mcp/memory/composio) · confirm-class | done | `console/overlays/router.py` `OVERLAY_HANDLERS` (12 entries → awaited `push_screen` flows); `overlays/pickers.py`, `sessions.py`, `auth.py` · `tests/console/test_overlays.py::TestRouterTable::test_covers_every_modal_kind` (`len(OVERLAY_HANDLERS)==12` vs `MODAL_KINDS`) |
+| Settings overlay rows | done | Python 15 rows == TS `pickers.tsx:418-542` (15). **Plan-doc erratum:** §6 says "16 rows"; TS ground truth has 15 — port matches TS, not the plan. |
+| 4 theme schemes w/ live preview | done | `console/theme/resolve.py` THEMES: midnight, daylight, midnight-cb, daylight-cb; preview-on-highlight w/ Esc-revert / Enter-commit-persist (`pickers.py::run_theme_picker`) · `tests/console/test_theme.py` (9 classes) |
+| 21-chord `/keys` + chord latch + double-Ctrl+C | done | HOTKEYS (21) `workbench.py`; `input/intents.py` INTENT_TABLE; `input/chord.py` `advance_chord` (600 ms) + `advance_exit_window` (500 ms) · `tests/console/test_input.py` (TestIntentTable 15 cases, TestAdvanceChord, TestExitWindow); live smoke (c) clean `/quit` |
+| Startup map + changelog splash keyed to `lastSeenVersion` | done | `console/startup.py` `gather_startup`/`gather_changelog` (full/condensed/none modes) · `tests/console/test_startup.py` |
+| Clipboard image paste + external `$EDITOR` | done | `kit/clipboard_image.py`, `kit/external_editor.py`, wired at `console/app.py:1225-1246` (`input:pasteImage`, `input:externalEditor`); paste markers via framework `EditorCore` · `test_input.py::TestPasteMarkers` |
+
+### 4. Channels — 5/5 done
+
+| Row | Status | Evidence |
+|---|---|---|
+| `SESSION_OPS` ×6 (`submit abort snapshot resume listModels cycleModel`) w/ flat `LinkSnapshot` | done | `channels/session_ops.py:163-172` registry; `project_snapshot` L63-104 (TS wire vocabulary verbatim); optional `cycleModel` probed à la TS `cycleModel?.()` · `tests/channels/test_channels.py:320-349`; live smoke (b) |
+| Dialog bridge `ask`/`tell` + 7 methods, 90 s timeout→fallback | done | `channels/link/dialog.py:187-216` frozen `DIALOG_TABLE` (select/confirm/input/editor=ask; notify/status/title=tell); `loop.call_later(dialogMs/1000)`; `dialogMs=90_000` `contract.py:544` ↔ TS `contract.ts:542-546` · linked-server round-trips `test_channels.py:297-316` |
+| Oneshot text + ndjson (start/signal/end frames, exit by phase) | done | `channels/oneshot.py:75-131`; exit 0/1 by `phase=="faulted"` L68-72 · `test_channels.py:240-289` (3 strategy tests); live smoke (a) exercised text path |
+| Framer U+2028/29 + split-rune reassembly | done | `channels/framer.py:64-89` explicit `
`/`
` escape (load-bearing — `json.dumps(ensure_ascii=False)` does **not** escape them); `codecs.getincrementaldecoder("utf-8")` fresh per call L92-122 · `test_channels.py:185-221` incl. mid-rune emoji split pin |
+| JSON-RPC codes exact + concurrent requests | done | `channels/contract.py:163-176` `OP_ERROR` = −32700/−32600/−32601/−32602/−32000; `link/server.py:191-217` per-request `asyncio.create_task` + inflight set + `gather` at stream end (deadlock-free asks) · `test_channels.py:340-349` |
+
+### 5. Addons + capability deck — 8/8 done
+
+| Row | Status | Evidence |
+|---|---|---|
+| 13 HookEvents × observe/transform/gate; gates only `tool:before`/`input:submit`/`compact:before` | done | `addons/contract.py:282-296` (13-tag union == TS `contract.ts:219-232`); EVENT_TRAITS ×13 dispatch table; RESERVED_EVENTS = the 3 gate-bearers · `tests/addons/test_addons.py:477-484` key-coverage test |
+| Interceptor onion (enter-forward, exit-reverse, error-recovery) | done | `AddonInterceptorChain` (tool interceptor): enter in order w/ block short-circuit, exit reversed, stage-throw → `AddonFault` no-op, exit may recover tool error · `test_addons.py:269-305` |
+| Addon commands/tools w/ reserved-name + duplicate conflict faults | done | `addons/host.py:447-460`; first-wins; `AddonFault(kind="conflict")` · `test_addons.py:399-429` |
+| `.indus/addons` `*.py` discovery, importlib loader, fault isolation | done | `addons/manifest.py` discovery walk (bare `*.py` + `__init__.py` dirs, dot-skip, sorted, never throws); `addons/loader.py` `_ImportlibLoader` via `spec_from_file_location` (fresh synthetic module name per load); load error → `AddonFault(kind="load")`, agent never crashes · real-importlib end-to-end `test_addons.py:492-540` |
+| Profiles `authoring`/`survey`/`all` w/ exact inverted membership | done | `capability_deck/manifest.py:109-126` PROFILE_TABLE verbatim incl. the TS inversion quirk (preserved per PLAN.md:164/259) · `tests/capability_deck/test_cards_provision.py` |
+| 12 framework built-ins | done | `builtin_bridge.py:307-310` `builtin_descriptors()`: read, ls, grep, find, websearch, webfetch, todo_read, write, edit, bash, process, todo_set (== TS `builtin-bridge.ts:124-233`; todo wire names follow the Python framework's `todo_read`/`todo_set`) |
+| 5 app cards (todo · bg-process real spawn/poll/stop · task · saas-action · memory) | done | `cards/__init__.py:110-116` `APP_NOVEL_CARDS`; DaemonTable on `asyncio.create_subprocess_shell`, 2000-line ring buffers, SIGTERM→SIGKILL 3 s grace · `test_cards_provision.py` spawns a **real** subprocess, polls, stops |
+| Bridge ledger (content/ULID keys, enroll/retire/withdraw, mount/detach `DeckFault("bridge")`) | done | `bridge_ledger` ledger (immutable log + pure `reduce_ledger` fold, upsert-by-key); sha256 content key / ULID; `DeckFault("bridge")` returned, never thrown · `tests/capability_deck/test_contract_ledger.py` |
+
+### 6. Sessions, transcripts, export, runtime-bridge, aux — 14/14 done
+
+| Row | Status | Evidence |
+|---|---|---|
+| Branchable NDJSON tree `indus/transcript@1` | done | `conductor/contract.py` `TRANSCRIPT_SCHEMA`; `transcript_store.py` parent-linked tree; `serialize.py` envelope · `tests/conductor/test_catalog_store.py`; live artifact header confirmed |
+| Resume by id + deepest-leaf recovery | done | `conductor.py` `resume()`; `_deepest_leaf` (ULID tie-break) · `test_catalog_store.py::test_deepest_leaf_*` |
+| Branch/fork from prior turn | done | `transcript_store.py` `branch_at()` (repoint leaf, no rewrite) · `tests/conductor/test_queue_fork.py` |
+| Tree navigation | done | `sessions/library.py` `tree()` + `prior_turns()` projections · `tests/sessions/test_sessions.py` |
+| Per-cwd `--<slug>--` + `--continue` | done | `boot/runners/session.py` `session_scope_dir` · `tests/boot/test_session_persist.py`; live artifact path confirmed |
+| Legacy framework-session import | done | `serialize.py` `import_legacy_file/_text`, `resolve_legacy_messages` → framework loaders (`load_entries_from_file`, `parse_session_entries`, …) · conductor legacy tests |
+| `/export` self-contained HTML (base64 payload, SGR-painted tool output, widget splice) | done | `transcript_export/publish.py` (base64 `data-turns`), `sgr.py` state machine (bold/red/256/24-bit), widget splice by toolCallId in `template.py` · `tests/transcript_export/test_transcript_export.py` |
+| `/share` gh-gist degradations | done | `integrations.py` `_run_share`: `gh gist create`, `FileNotFoundError`/auth-failure → local-path fallback message · `test_commands_integrations.py` |
+| Runtime-bridge 3 dialects (claude-cli, codex-cli, indusagi-cli) | done | `runtime_bridge/bridges/{claude_cli,codex_cli,indusagi_cli}.py` → provider-neutral `NormalizedEvent`; `sink.py` → framework `AssistantMessageEventStream` · `tests/runtime_bridge/test_runtime_bridge.py` (17 cases: sink 3, dialects 2+2+2, broker 8, credentials 4) |
+| Runtime-bridge wiring stance | done (parity note) | Library-only in **both** builds: "runtime-bridge is currently consumed only via the package barrel (`src/index.ts: export * as runtimeBridge`) — it is a complete, fully-tested library layer not yet wired into the boot path" (analysis 01 §1, line 12); Python mirrors via `runtime_bridge/__init__.py:14-16` |
+| Auth vault multi-account + OAuth refresh + default promotion, 0600 | done | `boot/auth_vault.py` `_SECURE_FILE_MODE=0o600` (chmod on every rewrite); refresh at 60 s expiry margin; legacy-shape tolerance · `tests/launch/test_launch.py` vault cases |
+| `primeProviderEnv`/key-resolver parity | done | `boot/runners/session.py` `prime_provider_env` (PROVIDER_ENV map) + `build_key_resolver` (`--account` → default → first) · `tests/boot/test_session_persist.py` |
+| Window-budget 0.75/6000/2048 + manual `/summarize-context` w/ guidance | done | `window_budget/condenser.py` `DEFAULT_POLICY(trigger_ratio=0.75, keep_recent=6000, reserve_tokens=2048)`; 3.6 chars/token (TS value, not framework 4.0); identity-on-under-budget asserted with `is`/`is not` · `tests/window_budget/test_window_budget.py` |
+| Insight wrapper round-trip (collector + replay) | done | `insight/` wraps `indusagi.tracing` (Probe→Segment vocabulary; `v:1` NDJSON + `[insight:scrubbed]` pinned in wrapper); collector sink + replay reader are the app-owned pieces · `tests/insight/test_insight.py` (round-trip, FNV-1a gate, redaction + 4096 truncation) |
+
+---
+
+## Deliberate waivers / documented divergences (all plan-cited)
+
+| # | Waiver | Citation |
+|---|---|---|
+| 1 | **Composer deleted** — TS software-caret `Composer.tsx` replaced by framework `PromptEditor` | PLAN.md §1 line 24, §2 line 104; analysis 02 §6 |
+| 2 | **Paste vault/burst-debounce deleted** — replaced by framework `EditorCore` paste markers + native `events.Paste` (`paste.ts` gone) | PLAN.md §1 line 24, §3 line 177; analysis 02 §6 |
+| 3 | **jiti → importlib** — addons are `.py`, loaded via `importlib.util.spec_from_file_location`; virtual-modules/alias bridge collapses, `BUNDLED_NAMESPACES` vestigial | analysis 03 lines 126, 153, 199; PLAN.md line 79 |
+| 4 | **insight wraps `indusagi.tracing`** — not re-ported (2.5k LOC stays in framework); only collector sink, replay reader, FNV-1a gate are app code | PLAN.md line 24 ("locked"), line 165; analysis 05 §3 |
+| 5 | **repl placeholder gone** — M0's `cli.py` stub runner (PLAN.md line 200-201) replaced in M5 by the real Textual console mount (`boot/runners/repl_runner.py` defaults to `induscode.console.mount.mount_console`; seam stays injectable for tests) | PLAN.md line 237 ("repl-runner wired") |
+| 6 | **runtime-bridge not wired into boot** — parity with TS, where it is also a barrel-only library layer | analysis 01 §1 line 12 (quoted verbatim above) |
+| 7 | **`/memory` cosmetic parity** — `on/off` flips a reporting flag only, never removes the memory card from the live deck; `# parity:` comment at `integrations.py:122-124` | PLAN.md lines 164, 259 (quirk preservation); analysis 03 line 196 |
+| 8 | **Two compaction engines stay distinct** — `window_budget` (tokens, 0.75) is the conductor's CondenseFn; framework `runtime.memory` (turns, 0.8) never wired in | PLAN.md lines 166, 255; analysis 05 §1/risk-3 |
+| 9 | **Chord-latch break semantics** — TS broke a primed double-tap on any key; the Textual editor consumes printables, so the latch is timer-bounded (600 ms window and outcomes identical); documented at `console/app.py` | analysis 02 (input collapse); divergence note in code |
+| 10 | **OAuth overlay redesigned on `asyncio.Future`** — parked-resolver instead of TS callback threading; flow integrity (provider threading, vault refresh) preserved | PLAN.md M5 line 237; analysis 02 risk-1 |
+| 11 | **TypeBox → literal JSON-schema dicts** — framework `parameters: Mapping` convention; runtime guards for `action` discriminants | PLAN.md line 176; analysis 03 §6.7 |
+| 12 | **Modal routing inversion** — TS always-mounted overlay components → awaited `push_screen` flows with typed outcomes; reducer modal bookkeeping preserved | PLAN.md M5 line 237 ("overlays as push_screen flows") |
+| 13 | **todo wire names** — Python framework exposes `todo_read`/`todo_set` (TS framework: `todoread`/`todowrite`); the deck honors the framework wire contract verbatim | builtin_bridge.py:307-310; PLAN.md framework-convention rule (line 176) |
+
+## Plan-doc errata (no port action needed)
+
+1. **§6 "settings (16 rows)"** — TS `pickers.tsx:418-542` defines **15** rows; Python matches TS (15). The plan overstated by one.
+2. **§6 `--tools` "(… `_`/`-`-insensitive)" attributed to `launch/invocation/flags.py`** — in both TS and Python the canonicalization lives in the boot session runner's deck filter (`session.ts:261` / `session.py:314-330`); the parser-level roster filter is exact-match in both. Behavior parity is exact, including the empty-allowlist→all-tools quirk.
+
+## Follow-ups (non-blocking)
+
+1. **[upstream / indusagi-python-rebuild]** Fix `_FacadeTranslator._on_delta` first-delta framing (root cause above) so the first streamed chunk reaches `MessageUpdateEvent`; then re-run smoke (a) expecting exactly `pong`. Add a framework regression test: first text delta must surface as a `text_delta` update.
+2. Once fixed upstream, add an induscode integration pin: oneshot `-p` output equals the full assistant text for a multi-chunk stream.
+3. Optional: dedicated per-method timeout test for the dialog bridge's 90 s fallback (currently covered via the linked-server round-trip suite).

induscode-0.1.0/PKG-INFO

@@ -0,0 +1,97 @@
+Metadata-Version: 2.4
+Name: induscode
+Version: 0.1.0
+Summary: Indusagi coding agent — terminal-first AI coding agent on the indusagi framework (Python rebuild)
+Project-URL: Homepage, https://github.com/varunisrani/indusagi-ts
+Project-URL: Repository, https://github.com/varunisrani/indusagi-ts
+Project-URL: Issues, https://github.com/varunisrani/indusagi-ts/issues
+Author: Varun Israni
+Author-email: IndusAGI Team <team@indusagi.ai>
+License-Expression: MIT
+License-File: CREDITS.md
+License-File: NOTICE
+Keywords: agent,ai,cli,coding-agent,llm,mcp,tui
+Classifier: Development Status :: 4 - Beta
+Classifier: Environment :: Console
+Classifier: Intended Audience :: Developers
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Software Development :: Code Generators
+Classifier: Typing :: Typed
+Requires-Python: >=3.11
+Requires-Dist: indusagi[mcp,tui]>=0.1.0
+Requires-Dist: markdown-it-py>=3.0
+Requires-Dist: pygments>=2.18
+Requires-Dist: python-ulid>=2.7
+Provides-Extra: dev
+Requires-Dist: pytest-asyncio>=0.24; extra == 'dev'
+Requires-Dist: pytest>=8; extra == 'dev'
+Description-Content-Type: text/markdown
+
+# induscode (Python rebuild)
+
+Terminal-first AI coding agent on the [indusagi framework](../indusagi-python-rebuild)
+— the Python port of the TypeScript `indusagi-coding-agent` (lineage v0.1.62).
+Console scripts: **`pindus`** (primary) and **`induscode`** (alias).
+
+Port plan: `../indus-code-rebuild/PYTHON_PORT_PLAN/PLAN.md` (milestones M0–M6).
+Current status: **all milestones built** — conductor turn loop, runtime bridge,
+capability deck, addons, headless CLI (launch/boot/channels/sessions), Textual
+console with 26 slash commands, transcript export, and packaging. **649 tests
+green**; lineage scan clean; wheel verified in a fresh venv.
+
+## Install (from wheels)
+
+The `indusagi` framework is a local dependency (not on PyPI), so install its
+wheel first — with the `[mcp,tui]` extras, which the agent requires — then the
+`induscode` wheel:
+
+```bash
+python3.13 -m venv agent-venv
+agent-venv/bin/pip install '../indusagi-python-rebuild/dist/indusagi-0.1.0-py3-none-any.whl[mcp,tui]'
+agent-venv/bin/pip install dist/induscode-0.1.0-py3-none-any.whl
+```
+
+Build the wheels with `python -m build` in each project if `dist/` is missing
+or stale.
+
+## Run
+
+```bash
+pindus                                  # interactive Textual console
+pindus -p "explain this repo"           # one-shot: print the result and exit
+pindus -p --json "explain this repo"    # headless NDJSON line protocol
+pindus --list-models                    # provider/model catalog
+pindus signin                           # store credentials (OAuth or API key)
+pindus signin --list                    # show stored accounts
+pindus --help                           # full flag table (@file attachments, --, etc.)
+```
+
+State lives in **`~/.pindusagi`** (shared with the framework; the
+`INDUSAGI_CODING_AGENT_DIR` then `INDUSAGI_HOME` environment variables
+override). Credentials are stored in the agent's multi-account vault
+(`auth.json`, mode 0600).
+
+## Dev setup
+
+```bash
+python3.13 -m venv .venv
+.venv/bin/pip install -e ../indusagi-python-rebuild -e ".[dev]"
+```
+
+## Gates
+
+```bash
+.venv/bin/pytest -q                            # test suite — 649 tests, never weakened
+.venv/bin/python scripts/lineage_scan.py       # clean-room lineage guard
+.venv/bin/pindus --version                     # smoke: prints "induscode 0.1.0"
+.venv/bin/python -m build                      # sdist + wheel into dist/
+```
+
+The packaging gate additionally installs both wheels into a throwaway venv and
+runs `--version`, `--help`, `--list-models`, and the subsystem imports.
+
+MIT © Varun Israni — see `NOTICE` and `CREDITS.md`.