release: canary → main (31 commits) — Python truth-layer + cross-platform install + 13 bug fixes

[joelteply]

What's in this bundle

31 commits since main, accumulated through canary cross-validation by the AI peer team (vhsm-d1f4, ideem-local-4bef, continuum-b69f, authenticator-fd63). 13 issues closed in the most recent session alone (#179, #132, #91, #163, #161, #145, #94, #97, #98, #99, #184, #142, plus the issue that filed #188).

Architecture: Python truth-layer (#152 Phase 0+1, PRs #166-#175)

`airc` (bash) is the user surface; business logic moves to `lib/airc_core/*` Python modules called via argparse CLIs:

• `airc_core.config` — config.json CRUD (`get`, `get_name`, `set_name`, `set_host_block`)
• `airc_core.datetime` — iso_to_epoch (BSD/GNU/python fallback unified)
• `airc_core.handshake` — pair handshake (joiner `send`, host `accept_one`, response `get_field`)
• `airc_core.monitor_formatter` — 250-line message stream formatter

Plus bash decomposition (`lib/airc_bash/*`) — `platform_adapters.sh`, `cmd_doctor.sh` extracted from the monolith. `airc` itself is now under 5000 lines (was ~5500).

All Python CLIs use argparse `--flags` for paths (per #174), not env vars — so MSYS path-translation on Git Bash is per-arg-predictable. `airc_core` carries the canonical types; bash callers are thin dispatchers.

Cross-machine substrate fixes (PRs #160, #164, #176-#178)

Cross-Mac/Windows airc messaging now verified end-to-end:

• fix(install,doctor): Windows HNS port-22 reservation (continuum's diagnosis — sshd bind EPERM) #160 — Windows HNS port-22 reservation (sshd bind EPERM under Hyper-V)
• fix(python): AIRC_PYTHON env var replaces broken export -f shim (THE root cause) #164 — `AIRC_PYTHON` env var replaces broken `export -f` shim (the root cause of "command not found" cascades)
• fix(airc_core): use argparse --flags for all paths, not env vars (continuum's MSYS catch) #174 — `airc_core.*` use argparse --flags for all paths, not env vars (MSYS path mangling fix)
• fix(airc_core): config set_host_block — close last env-var-pass site (continuum's #174 follow-up) #176 — config set_host_block closes the last env-var-pass site
• fix(msys): MSYS2_ARG_CONV_EXCL — last layer; cross-machine VERIFIED end-to-end #177 — `MSYS2_ARG_CONV_EXCL=/Users/;/home/;/root/` covers the last layer
• fix(encoding): PYTHONIOENCODING=utf-8 at airc startup (continuum's silent-drop catch) #178 — `PYTHONIOENCODING=utf-8` at airc startup (continuum's silent-drop / Latin-1 mojibake catch)

Windows install path fixes (PR #187 — closes #94, #97, #98, #99)

A Windows user can now do a clean install end-to-end:

• bug: install.ps1 — Tailscale winget package ID typo (lowercase) causes silent install failure #94 — Tailscale winget package id case (`Tailscale.Tailscale`); winget --exact is case-sensitive
• bug: install.ps1 — DefaultShell unconfigured; Windows airc message delivery fails (cmd.exe lacks 'cat') #98 — `install.ps1` configures `HKLM:\SOFTWARE\OpenSSH\DefaultShell` to Git for Windows bash
• bug: Get-RemoteHome — Windows-style host path gets mangled in SSH pipeline; one-line forward-slash fix #99 — `Get-RemoteHome` converts backslashes to forward slashes
• Windows host: backslash paths in SSH remote commands break under bash DefaultShell #97 — same root cause as bug: Get-RemoteHome — Windows-style host path gets mangled in SSH pipeline; one-line forward-slash fix #99 (backslash paths under bash DefaultShell)
• Plus: `tailscale status` and `gh auth status` no longer leak `$LASTEXITCODE` through to script natural-end exit

Stability fixes

• Pair-listener Python orphans survive parent bash death — port held + key auth continues silently #132 (PR fix(airc): pair-listener parent-watch — orphan-process port-hold (#132) #185) — pair-listener parent-watch: when airc bash dies, accept-loop subshell + python listener exit cleanly. Fixes 2 of 3 long-standing integration test flakes ("port still held after teardown" + "alpha still listening")
• airc nick rename doesn't propagate to sidecar scope (skill promises shared nick across rooms) #179 (PR fix(airc): rename propagates to sibling scopes; cmd_send --internal flag (#179) #183) — rename propagates to all sibling scopes (primary + sidecars), `cmd_send --internal` flag for informational broadcasts
• airc join: silent monitor exit on idle channel — ESCALATE_AFTER assumes daemon respawn #184 (PR fix(airc): surface monitor-escalation to stdout + daemon-aware (#184) #189) — silent monitor exit: escalation banner now goes to stdout + stderr, daemon-aware messaging
• airc list shows stale 1:1 invites with no auto-prune #142 (PR fix(airc list): hide stale entries by default; --all to show; --prune (#142) #190) — `airc list` hides stale entries by default; `--all` to show; `--prune` to clean up

CI infrastructure (PRs #186, #187)

`.github/workflows/ci.yml` — five jobs on every PR + push:

• clean-install-linux: ubuntu install.sh + airc doctor + smoke
• clean-install-macos: macos install.sh + airc doctor + smoke
• clean-install-windows: windows install.ps1 (pwsh) + airc doctor
• clean-install-windows-ps5: install.ps1 under Windows PowerShell 5.1 (the default Windows shell — guards bootstrap-airc.ps1 fails under PowerShell 5.1 (airc.ps1 requires 7+) #91)
• integration-suite: full `test/integration.sh` on push to canary/main

`AIRC_SKIP_PREREQS` is NOT used — the CI tests the real install path on stock runners. `CI=true` auto-detection in install.sh skips sshd setup (osascript admin prompt hangs in CI) + tailscale install (slow, optional, no tailnet).

Smaller fixes

• fix(scope): _primary_scope_for empty-prefix guard (vhsm e2e catch on PR #144) #150 — _primary_scope_for empty-prefix guard (vhsm e2e catch)
• fix(prereqs): strict python3 probe — Windows Store alias defeats command -v #153 — strict python3 probe (Windows Store alias defeats command -v)
• fix(sidecar): inherit --no-gist flag so tests stop leaking live #general gists #154 — sidecar inherits --no-gist flag (Test fixtures leak into live #general (alpha joins polluting room) #145 fix)
• fix(gist): git-clone fallback + |\| true guards under rate-limit (this limit will kill people) #155 — gist git-clone fallback under rate-limit
• feat(doctor,install): probe sshd readiness + scope strict-probe to python (PR #153 follow-up) #156, feat(install): auto-install + start sshd during install (close "this needs to be in the install" gap) #157 — sshd readiness probe + auto-install + start
• fix(sshd-probe): macOS launchctl print system + osascript admin dialog (PR #156/#157 live-test followups) #158 — macOS sshd setup via launchctl + osascript admin dialog
• fix(doctor): tailscale probe uses resolve_tailscale_bin (catch GUI install) #159 — doctor tailscale probe uses resolve_tailscale_bin
• fix(parser,prereq): jq is required — fallback parser corrupts gist envelopes without it (continuum's diagnosis) #162 — jq required, fallback parser corruption (continuum's diagnosis)
• fix(airc): two PR #164 followups — sed missed line 1372 + harden host_* config write (continuum's retest) #165 — PR fix(python): AIRC_PYTHON env var replaces broken export -f shim (THE root cause) #164 followups (sed missed line + harden host_* write)

Risk profile

• Python is now a hard prereq (`airc_core` modules invoked at startup). Already documented; doctor + install.sh + install.ps1 all install it.
• The bash file got smaller (5500 → ~4900 lines via airc_core extraction); behavior should be identical.
• All 31 commits passed canary cross-validation by the AI peer team running them daily. Several were caught + fixed mid-stream (e.g. continuum-b69f's MSYS path-mangling diagnosis closed three commits later).

Test plan

• CI clean-install matrix passing on linux + macos + windows + windows-ps5
• Integration suite up from 180 → 183 passing on canary (two flakes resolved by Pair-listener Python orphans survive parent bash death — port held + key auth continues silently #132)
• Cross-machine Mac↔Windows verified end-to-end after fix(encoding): PYTHONIOENCODING=utf-8 at airc startup (continuum's silent-drop catch) #178
• Final full integration suite run on canary HEAD (the integration-suite CI job will run on push when this merges)

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This release bundle promotes canary → main by introducing a Python “truth-layer” (lib/airc_core/*), refactoring bash into sourced modules (lib/airc_bash/*), and hardening cross-platform install/CI—especially for Windows end-to-end setup and cross-machine messaging reliability.

Changes:

• Added airc_core Python CLIs (config CRUD, handshake, datetime, monitor formatter) and wired airc to invoke them via AIRC_PYTHON + PYTHONPATH.
• Extracted bash adapters/doctor into lib/airc_bash/* and updated integration tests accordingly.
• Added/updated install paths (bash + PowerShell), plus a multi-OS clean-install CI matrix and integration-suite gating.

Reviewed changes

Copilot reviewed 13 out of 14 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
airc	Switches to AIRC_PYTHON, sets MSYS/encoding env, sources extracted bash libs, and routes core logic through airc_core CLIs.
lib/airc_core/config.py	New argparse-driven config.json read/write helpers (get, get_name, set_name, set_host_block).
lib/airc_core/datetime.py	New Python ISO→epoch conversion used by bash adapters.
lib/airc_core/handshake.py	New joiner/host handshake implementation + field extraction CLI.
lib/airc_core/monitor_formatter.py	New Python monitor formatter module replacing embedded heredoc.
lib/airc_core/__init__.py	Introduces airc_core package metadata.
lib/airc_bash/platform_adapters.sh	Extracted cross-platform adapters; iso_to_epoch now delegates to Python.
lib/airc_bash/cmd_doctor.sh	Extracted airc doctor with expanded probes (jq/sshd/tailscale path resolution).
test/integration.sh	Updates platform adapter sourcing and adds deterministic iso_to_epoch assertions.
install.sh	Adds jq prereq + sshd enablement and CI-specific skips; updates prereq probing.
install.ps1	Adds jq, OpenSSH Server install/start, DefaultShell→bash, and explicit successful exit.
airc.ps1	Fixes winget Tailcale ID casing, remote-home path slash normalization, and $LASTEXITCODE leak in doctor.
.github/workflows/ci.yml	Adds clean-install matrix (linux/macos/windows/powershell5) + push-only integration suite.
.gitignore	Ignores Python bytecode artifacts.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Set-HnsPortFreedomFor22 uses reg add ... /d 0 without specifying the value type. If EnableExcludedPortRange is missing, reg add defaults to creating a REG_SZ, which may not be honored where a DWORD is expected. Prefer setting it via PowerShell with an explicit DWORD type (e.g., New-ItemProperty/Set-ItemProperty with -Type DWord) or add /t REG_DWORD to the reg.exe call.

Suggested change

	& reg add 'HKLM\SYSTEM\CurrentControlSet\Services\hns\State' /v 'EnableExcludedPortRange' /d 0 /f 2>$null | Out-Null
	& reg add 'HKLM\SYSTEM\CurrentControlSet\Services\hns\State' /v 'EnableExcludedPortRange' /t REG_DWORD /d 0 /f 2>$null | Out-Null

[Copilot]

In _airc_resolve_lib_dir(), the first candidate is "${AIRC_DIR:-}/lib". When AIRC_DIR is unset, that expands to "/lib", which is a real system directory and could (in rare cases) cause you to pick up an unrelated /lib/airc_core if present, or at least do an unnecessary filesystem probe. Consider only adding the AIRC_DIR-derived candidate when AIRC_DIR is non-empty (or default it to the intended install dir like $HOME/.airc-src).

Suggested change

	for _candidate in \
	"${AIRC_DIR:-}/lib" \
	if [ -n "${AIRC_DIR:-}" ]; then
	_candidate="$AIRC_DIR/lib"
	if [ -d "$_candidate/airc_core" ]; then
	# Canonicalize to absolute path so PYTHONPATH stays valid even
	# if cwd changes mid-script (heredocs that cd elsewhere). cd +
	# pwd is the portable canonicalize idiom — `realpath` and
	# `readlink -f` are not available everywhere (BSD readlink
	# lacks -f, busybox lacks realpath).
	_abs=$(cd "$_candidate" 2>/dev/null && pwd) || _abs="$_candidate"
	printf '%s' "$_abs"
	return 0
	fi
	fi
	for _candidate in \

[Copilot]

Workflow header comment says “Three jobs on every PR + every push”, but the workflow now defines four clean-install jobs (linux/macos/windows/windows-ps5) plus integration-suite on push. Update the comment block to match the actual job set so future edits don’t drift based on incorrect docs.

Suggested change

	# Three jobs on every PR + every push to canary/main:
	#
	# clean-install-linux ubuntu install.sh + airc doctor + smoke test
	# clean-install-macos macos install.sh + airc doctor + smoke test
	# clean-install-windows windows install.ps1 + airc doctor (PS-side)
	# Four clean-install jobs on every PR + every push to canary/main:
	#
	# clean-install-linux ubuntu install.sh + airc doctor + smoke test
	# clean-install-macos macos install.sh + airc doctor + smoke test
	# clean-install-windows windows install.ps1 + airc doctor (PS-side)
	# clean-install-windows-ps5 windows install.ps1 + airc doctor (PS 5.1)

[Copilot]

ensure_prereqs() is using a strict --version probe for all prereqs. Several of these tools (notably ssh-keygen on macOS/BSD) do not support --version and will exit non-zero even when correctly installed, which will make install.sh incorrectly treat them as missing and try to reinstall/fail. Consider limiting the strict probe to python/python3 (the Windows Store stub case) and using command -v (or a tool-appropriate version flag) for the rest.

Suggested change

	# Strict probe: presence on PATH AND a successful --version invocation.
	# The bare `command -v` form is fooled by Windows's Microsoft Store
	# python3.exe alias (continuum-b69f, 2026-04-27) — the file exists,
	# satisfies command -v, but exits 49 with a Store-redirect message
	# when actually run. Pre-fix: install printed "All required prereqs
	# present" and airc later silent-fail-cascaded at every python3 -c
	# invocation. Strict probe catches this at install time.
	if ! command -v "$cmd" >/dev/null 2>&1 || ! "$cmd" --version >/dev/null 2>&1; then
	# Only python3 gets a strict execution probe. The bare `command -v`
	# form is fooled by Windows's Microsoft Store python3.exe alias
	# (continuum-b69f, 2026-04-27) — the file exists, satisfies
	# command -v, but exits 49 with a Store-redirect message when
	# actually run. Other tools are checked for presence on PATH only,
	# because some valid platform variants (notably ssh-keygen on
	# macOS/BSD) do not support `--version`.
	local cmd_missing=0
	case "$cmd" in
	python3)
	if ! command -v "$cmd" >/dev/null 2>&1 || ! "$cmd" --version >/dev/null 2>&1; then
	cmd_missing=1
	fi
	;;
	*)
	if ! command -v "$cmd" >/dev/null 2>&1; then
	cmd_missing=1
	fi
	;;
	esac
	if [ "$cmd_missing" -eq 1 ]; then

[Copilot]

monitor_formatter arms the watchdog (SIGALRM handler / Timer thread) at import time. This means any import of airc_core.monitor_formatter in another module or in tests will mutate global process signal handlers / spawn a timer thread, which is a surprising side effect and can break embedding. Consider moving watchdog setup into run() (and teardown on exit) so importing the module is side-effect-free.

[Copilot]

_rename_files() returns True even if the rename/update fails (exceptions are swallowed and the function unconditionally returns True once the old file exists). That can cause _handle_rename() to print “nick old → new” and suppress further processing even though the peer files were not actually renamed/updated. Return False when os.rename/JSON update fails, and avoid swallowing exceptions that leave the peer directory in a partially-updated state.

Suggested change

	try:
	os.rename(old_json, new_json)
	d = json.load(open(new_json))
	d["name"] = new
	json.dump(d, open(new_json, "w"), indent=2)
	except Exception:
	pass
	old_pub = os.path.join(peers_dir, f"{old}.pub")
	new_pub = os.path.join(peers_dir, f"{new}.pub")
	if os.path.isfile(old_pub):
	try:
	os.rename(old_pub, new_pub)
	except Exception:
	pass
	return True
	old_pub = os.path.join(peers_dir, f"{old}.pub")
	new_pub = os.path.join(peers_dir, f"{new}.pub")
	tmp_json = f"{new_json}.tmp"
	original = None
	json_renamed = False
	pub_renamed = False
	try:
	with open(old_json) as f:
	original = json.load(f)
	os.rename(old_json, new_json)
	json_renamed = True
	updated = dict(original)
	updated["name"] = new
	with open(tmp_json, "w") as f:
	json.dump(updated, f, indent=2)
	os.replace(tmp_json, new_json)
	if os.path.isfile(old_pub):
	os.rename(old_pub, new_pub)
	pub_renamed = True
	return True
	except Exception:
	if os.path.exists(tmp_json):
	os.unlink(tmp_json)
	try:
	if pub_renamed and os.path.isfile(new_pub):
	os.rename(new_pub, old_pub)
	if json_renamed and os.path.isfile(new_json):
	os.rename(new_json, old_json)
	if original is not None and os.path.isfile(old_json):
	with open(tmp_json, "w") as f:
	json.dump(original, f, indent=2)
	os.replace(tmp_json, old_json)
	except Exception:
	if os.path.exists(tmp_json):
	os.unlink(tmp_json)
	raise
	return False

[Copilot]

accept_one writes the joiner-provided ssh_pub directly into ~/.ssh/authorized_keys after only a .strip(). If ssh_pub contains embedded newlines, this will append multiple lines/options and can unintentionally grant additional access beyond a single key. Validate that the value is exactly one line and matches an expected authorized_keys public-key format before appending (and consider rejecting/closing the connection on parse errors).

[Copilot]

Pull request overview

Copilot reviewed 14 out of 15 changed files in this pull request and generated 6 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

ensure_prereqs uses a blanket "$cmd" --version strict-probe for all prereqs (git/gh/jq/openssl/ssh-keygen/python3). Several of these do not support --version (notably ssh-keygen on macOS/BSD; and openssl commonly uses openssl version), so this will falsely mark them missing and trigger unnecessary/incorrect installs. Suggest probing only python/python3 with --version (to defeat the Windows Store stub), and using command -v (or command-specific version flags) for the rest.

Suggested change

	# Strict probe: presence on PATH AND a successful --version invocation.
	# The bare `command -v` form is fooled by Windows's Microsoft Store
	# python3.exe alias (continuum-b69f, 2026-04-27) — the file exists,
	# satisfies command -v, but exits 49 with a Store-redirect message
	# when actually run. Pre-fix: install printed "All required prereqs
	# present" and airc later silent-fail-cascaded at every python3 -c
	# invocation. Strict probe catches this at install time.
	if ! command -v "$cmd" >/dev/null 2>&1 || ! "$cmd" --version >/dev/null 2>&1; then
	# Probe python3 more strictly: the bare `command -v` form is fooled
	# by Windows's Microsoft Store python3.exe alias
	# (continuum-b69f, 2026-04-27) — the file exists, satisfies
	# command -v, but exits 49 with a Store-redirect message when
	# actually run. For the other prereqs, command-specific `--version`
	# handling is not portable (e.g. ssh-keygen on macOS/BSD, openssl
	# typically uses `openssl version`), so only require PATH presence.
	if ! command -v "$cmd" >/dev/null 2>&1 || { [ "$cmd" = "python3" ] && ! "$cmd" --version >/dev/null 2>&1; }; then

[Copilot]

The elevated Windows sshd setup passes a multi-line PowerShell script via -ArgumentList '-NoProfile -Command "$_elevated_payload"', but _elevated_payload contains many double quotes. As expanded, those quotes will break the nested -Command "..." argument parsing, so the elevated process is very likely to fail or run partial code. Use a safer transport (e.g., write the payload to a temp .ps1 file and Start-Process powershell.exe -File <path>, or use -EncodedCommand) so quoting/newlines are preserved.

[Copilot]

accept_one uses jname = joiner["name"] directly in filenames (os.path.join(peers_dir, jname + ".json")). Since this listener is reachable over the network, a malicious joiner could send a name containing path separators (e.g. ../...) and write outside peers_dir. Please validate/sanitize jname (e.g., enforce the same [a-z0-9-]+ constraint used elsewhere) and reject/close the connection on invalid input before doing any file writes.

[Copilot]

accept_one reads from the socket into data until it sees a newline, but there is no maximum size guard. A client can stream arbitrary bytes without \n, causing unbounded memory growth (DoS). Consider enforcing a max payload size (e.g., stop reading and close the connection after N bytes) and handling JSON decode errors gracefully.

[Copilot]

iso_to_epoch uses PEP 604 union syntax (int | None), which requires Python 3.10+. The repo doesn’t appear to document a minimum Python version, and some Linux distros still default to <3.10. Either document/enforce a 3.10+ minimum in install/docs, or switch to Optional[int] / int | None-free typing for wider compatibility.

Suggested change

	def iso_to_epoch(ts: str) -> int | None:
	from typing import Optional
	def iso_to_epoch(ts: str) -> Optional[int]:

[Copilot]

cmd_set_name / cmd_set_host_block docstrings say they write config "atomically", but the implementation writes directly to args.config (json.dump(..., open(args.config, "w"))). If the process is interrupted mid-write, config.json can be left truncated/corrupt. Consider writing to a temp file in the same directory and os.replace() it into place, and use with open(...) to avoid leaking file descriptors.

[Copilot]

Pull request overview

Copilot reviewed 14 out of 15 changed files in this pull request and generated 5 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The elevated PowerShell payload is interpolated directly into a -Command "$_elevated_payload" argument. Because _elevated_payload contains many embedded double quotes and newlines, this is very likely to break command-line parsing (and can fail unpredictably) on Windows Git Bash. Safer pattern: write the payload to a temp .ps1 file and Start-Process powershell.exe -File <temp> (or use -EncodedCommand) so quoting/newlines don’t corrupt the script content.

[Copilot]

cmd_set_host_block also writes config via json.dump(..., open(args.config, "w")) despite the docstring stating it’s atomic. Same risk of partial writes/corruption on interruption; recommend temp-file + os.replace() and with open(...) for both read/write handles.

[Copilot]

cmd_send builds payload using json.loads(args.my_identity_json ...) outside the socket try/except. If --my-identity-json is malformed, this will raise and crash without a clear error, bypassing the existing error handling. Suggest validating/parsing the JSON in a try/except and returning a non-zero exit with a helpful stderr message.

Suggested change

	payload = json.dumps({
	"name": args.my_name,
	"host": args.my_host,
	"ssh_pub": args.my_ssh_pub,
	"sign_pub": args.my_sign_pub,
	"airc_home": args.my_airc_home,
	"identity": json.loads(args.my_identity_json or "{}"),
	try:
	identity = json.loads(args.my_identity_json or "{}")
	except (ValueError, TypeError) as e:
	print(f"airc-handshake-send-error: invalid --my-identity-json: {e}", file=sys.stderr)
	return 1
	payload = json.dumps({
	"name": args.my_name,
	"host": args.my_host,
	"ssh_pub": args.my_ssh_pub,
	"sign_pub": args.my_sign_pub,
	"airc_home": args.my_airc_home,
	"identity": identity,

[Copilot]

cmd_accept_one does joiner = json.loads(data.decode().strip()) without any validation/exception handling. A malformed/partial handshake line will crash the listener (DoS) and may leave the connection/socket unclosed. Recommend wrapping decode/parse in try/except, closing conn, and returning cleanly (or emitting a structured error response) on bad input.

Suggested change

	joiner = json.loads(data.decode().strip())
	try:
	payload = data.decode().strip()
	joiner = json.loads(payload)
	if not isinstance(joiner, dict):
	raise ValueError("handshake payload must be a JSON object")
	except (UnicodeDecodeError, json.JSONDecodeError, TypeError, ValueError) as e:
	print(f"airc-handshake-accept-error: invalid handshake payload: {e}", file=sys.stderr)
	conn.close()
	sock.close()
	return 1

[Copilot]

_rename_files (and _find_peer_by_host) uses json.load(open(...)) / json.dump(..., open(...)) without context managers, which can leak file descriptors in a long-running monitor (renames may happen repeatedly). Use with open(...) as f: for both reading and writing, and consider not swallowing all exceptions silently here to aid debugging when a rename fails.

[Copilot]

Pull request overview

Copilot reviewed 13 out of 15 changed files in this pull request and generated 5 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

In the staged elevated PowerShell payload, the reg add ... EnableExcludedPortRange line omits /t REG_DWORD. reg add defaults to REG_SZ, which can set the wrong registry type and make the HNS workaround ineffective. Update the payload (and the printed manual-fix snippet) to specify the value type explicitly.

Suggested change

	reg add "HKLM\SYSTEM\CurrentControlSet\Services\hns\State" /v "EnableExcludedPortRange" /d 0 /f | Out-Null;
	reg add "HKLM\SYSTEM\CurrentControlSet\Services\hns\State" /v "EnableExcludedPortRange" /t REG_DWORD /d 0 /f | Out-Null;

[Copilot]

The Windows sshd manual fix hint uses reg add ... EnableExcludedPortRange /d 0 without specifying /t REG_DWORD. Since reg add defaults to REG_SZ, this guidance may lead to the wrong registry value type. Consider updating the printed command to include /t REG_DWORD so the workaround reliably applies.

Suggested change

	printf " reg add HKLM\\\\SYSTEM\\\\CurrentControlSet\\\\Services\\\\hns\\\\State /v EnableExcludedPortRange /d 0 /f\n"
	printf " reg add HKLM\\\\SYSTEM\\\\CurrentControlSet\\\\Services\\\\hns\\\\State /v EnableExcludedPortRange /t REG_DWORD /d 0 /f\n"

[Copilot]

The docstring says this write is "Atomically" performed, but the implementation writes directly to args.config via json.dump(..., open(..., "w")), which can truncate/corrupt the file if interrupted mid-write. Either implement an atomic write (write to a temp file + os.replace) or adjust the docstring to avoid claiming atomicity.

[Copilot]

This docstring claims the host_* block write is "Atomically" performed, but the function dumps JSON directly to the target path. To match the contract, consider writing to a temporary file and os.replace() it into place; otherwise, update the wording to reflect that it's a normal write.

[Copilot]

In _adapter_call, the inner bash -c sets AIRC_PYTHON using single quotes (export AIRC_PYTHON='${AIRC_PYTHON:-python3}'), which assigns the literal string ${AIRC_PYTHON:-python3} rather than expanding it. This will break adapters that rely on $AIRC_PYTHON (e.g. iso_to_epoch) on platforms where AIRC_PYTHON is not exactly python3. Prefer relying on the env var passed to bash -c, or export it without single quotes.

Suggested change

	bash -c "source '$_adapters_file'; export AIRC_PYTHON='${AIRC_PYTHON:-python3}'; $*"
	bash -c "source '$_adapters_file'; $*"