canary → main: identity layer, derive_name fix, IRC parity docs

[joelteply]

Promote canary → main. Six commits across four PRs since the last promote (#68).

What's landing

PR	summary
#67	skills(join): align with auto-scope + narrate-events guidance — fixes stale #general references; adds §2a identity bootstrap; adds §2b mandatory event-narration; adds canary/update troubleshooting bullet.
#69	fix(derive_name): bump basename cap from 12 to 20 chars — closes the authenticato-fd63 truncation bug visible to fresh-clone users on main today. Sanitization tightened (collapse double-dashes, strip leading/trailing).
#70	feat(identity): agent identity layer with cross-platform linking (#34) — pronouns/role/bio/status/integrations on top of derive_name, exchanged at pair-handshake; new airc identity show/set/link/import/push, airc whois, airc kick. Backward-compat with existing peers (missing fields default to {}).
#71	docs(readme): cover the new identity layer — IRC-vs-airc table additions, Magic bullet, Core Commands subsection, full "Agent identity & WHOIS" section.
#72	chore: drop WINDOWS-PORT-STATUS.md — port shipped, tracker no longer load-bearing.

Plus a load-bearing test-harness fix folded into #70: cleanup_dirs was silently broken on macOS because find /tmp -maxdepth 1 doesn't traverse the /tmp → /private/tmp symlink. Fixed by resolving to canonical path before walking. Linux unaffected.

Sign-off status

	sign-off	notes
Mac (Joel's machine)	✅	126/126 integration tests pass on canary HEAD; live commands all working in #useideem host scope
Windows (green via airc DM)	⏳ pending	DM'd twice (post-#70 merge + post-canary blessing); no reply yet. Joel chose to ship without explicit Windows blessing — additions are backward-compat (PS1 joiner pairing with bash host gets default {} identity) so the risk is asymmetric feature-set, not breakage.

Out of scope (already tracked, not blocking)

• airc.ps1 parity for the new identity / whois / kick commands (coordinating with green)
• airc ban (permanent block, complement to kick)
• Live slack / telegram / discord import-push (continuum is the v1 integration that actually works)

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Promotes canary → main by landing the new identity layer (handshake exchange + CLI), fixing derive_name truncation/sanitization, and updating docs/skills for IRC-parity + auto-scope guidance.

Changes:

• Add identity management commands (airc identity ...), airc whois, and airc kick, including handshake-time identity exchange.
• Improve derive_name (basename cap 12 → 20, tighter dash sanitization) and preserve identity when reconnecting by merging config.json.
• Extend integration tests + update README and join skill docs; fix macOS /tmp cleanup in the test harness.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
airc	Implements identity layer, whois/kick, handshake identity caching, and derive_name update.
test/integration.sh	Fixes /tmp cleanup on macOS; adds new identity/whois/kick scenarios.
skills/join/SKILL.md	Updates join skill for auto-scope defaults + identity bootstrap + event narration guidance.
README.md	Documents auto-scope and the new identity/WHOIS/kick command set.
WINDOWS-PORT-STATUS.md	Removes obsolete Windows port tracking document.

---

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

[Copilot]

airc identity link supports unlinking by omitting/blanking the handle, but the usage/error text still implies <handle> is required. Update the usage string (and the cmd_identity help line) to reflect that the handle is optional and that an empty handle unlinks, so users aren't misled.

Suggested change

	[ -z "$platform" ] && die "Usage: airc identity link <platform> <handle>"
	[ -z "$platform" ] && die "Usage: airc identity link <platform> [handle] (omit or pass empty handle to unlink)"

[Copilot]

cmd_kick attempts to remove the peer's SSH key from ~/.ssh/authorized_keys using $PEERS_DIR/$target.pub, but the host handshake writes peers/<name>.pub from sign_pub (the signing key), not ssh_pub. As a result, kick will usually leave the joiner's SSH key authorized, so the peer may still be able to SSH even after being “kicked”. Store the joiner’s ssh_pub in the peer record (e.g. in <peer>.json) and use that value for authorized_keys removal, and adjust the confirmation message accordingly.

[Copilot]

scenario_kick's header comment says kick removes the peer's pubkey from authorized_keys, but the scenario never asserts that the joiner’s SSH public key was actually removed from the host’s ~/.ssh/authorized_keys. Adding an explicit before/after grep for the joiner’s ssh_key.pub would both cover the advertised behavior and catch regressions in the key-removal implementation.

[Copilot]

cmd_identity shifts two arguments after reading the subcommand (shift 2), which drops the first flag for set/link/import/push (e.g., airc identity set --pronouns they becomes args=["they"] and fails parsing). Change this to shift only the subcommand (or conditionally shift based on arg count) so flags are preserved.

[Copilot]

cmd_whois uses the user-supplied target directly in local/remote file paths (e.g. peer_file="$PEERS_DIR/$target.json" and cat $host_airc_home/peers/$target.json). This allows path traversal (e.g. ../config) and, in the remote case, shell injection if target contains metacharacters. Validate target against the same safe charset as nicknames (e.g. ^[a-z0-9-]+$) and/or properly quote/escape the remote path before passing it to SSH.

[Copilot]

cmd_kick constructs peer_file="$PEERS_DIR/$target.json" from the unsanitized CLI argument. A value like ../config would match an existing file and then be deleted by rm -f, which is an easy footgun and a path traversal vulnerability. Reject peer names that don't match the allowed nickname format (e.g. ^[a-z0-9-]+$) before using them in filesystem paths.

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 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]

The remote whois path builds an SSH command with an unquoted $host_airc_home (from config/handshake) embedded into the remote shell command. If the path contains spaces or shell metacharacters, this will fail; worse, a malicious/compromised host could inject additional shell syntax via host_airc_home. Consider robust quoting/escaping (or avoid the remote shell entirely by invoking a remote python/json reader with a safely-passed file path) before concatenating into an SSH command string.

Suggested change

	local remote_blob
	remote_blob=$(relay_ssh "$host_target" "cat $host_airc_home/peers/$target.json 2>/dev/null" 2>/dev/null || true)
	local remote_blob remote_peer_file remote_peer_file_q
	remote_peer_file="$host_airc_home/peers/$target.json"
	remote_peer_file_q=$(python3 -c '
	import shlex, sys
	print(shlex.quote(sys.argv[1]))
	' "$remote_peer_file" 2>/dev/null)
	remote_blob=$(relay_ssh "$host_target" "cat -- $remote_peer_file_q 2>/dev/null" 2>/dev/null || true)

[Copilot]

In the continuum import merge, ints["continuum"] is set to src.get("name", "") even when JSON parsing fails (src becomes {}), which overwrites any existing continuum link with an empty string. It would be safer to only update the integrations mapping when the imported persona name is non-empty (or fall back to the requested id), otherwise keep the existing link intact.

[Copilot]

This nick sanitization regression check can false-pass because the airc nick invocation ignores failures (|| true). If the command errors for an unrelated reason, $renamed may remain the old valid name and the test still passes. Consider asserting airc nick exits 0 and/or that the stored name actually changed to the expected sanitized value before checking the leading-dash invariant.

Suggested change

	AIRC_HOME="$home/state" "$AIRC" nick ".dottyname" >/dev/null 2>&1 || true
	local renamed; renamed=$(python3 -c "import json; print(json.load(open('$home/state/config.json')).get('name',''))" 2>/dev/null)
	case "$renamed" in
	-*) fail "airc nick produced leading-dash name '$renamed' — sanitization regression" ;;
	"") fail "airc nick wrote empty name — sanitization regression" ;;
	*) pass "airc nick strips leading dash post-sanitization (got '$renamed')" ;;
	esac
	local renamed
	if AIRC_HOME="$home/state" "$AIRC" nick ".dottyname" >/dev/null 2>&1; then
	renamed=$(python3 -c "import json; print(json.load(open('$home/state/config.json')).get('name',''))" 2>/dev/null)
	case "$renamed" in
	-*) fail "airc nick produced leading-dash name '$renamed' — sanitization regression" ;;
	"") fail "airc nick wrote empty name — sanitization regression" ;;
	"dottyname") pass "airc nick strips leading dash post-sanitization (got '$renamed')" ;;
	*) fail "airc nick wrote unexpected sanitized name '$renamed' (expected 'dottyname')" ;;
	esac
	else
	fail "airc nick .dottyname failed unexpectedly"
	fi

[Copilot]

This bullet still claims airc join auto-joins #general, but the default behavior described immediately below is now auto-scope to the current repo org (with #general only as a fallback). Updating this line to match the new default (or explicitly qualify it as the non-git-dir fallback case) would prevent reader confusion.

[Copilot]

Config writes here use an in-place json.dump(..., open(CONFIG, "w")), which can transiently leave a partially-written config.json if another process reads mid-write (the integration test later adds a sleep to avoid flaking on half-written config). Consider switching to an atomic write (write to a temp file, fsync, then rename/replace) so readers either see the old or new full JSON.

Suggested change

	import json, os
	try:
	c = json.load(open(os.environ["CONFIG"]))
	except Exception:
	c = {}
	c["name"] = os.environ["MY_NAME"]
	c["host"] = os.environ["MY_HOST"]
	c["host_target"] = os.environ["SSH_TARGET"]
	c["created"] = os.environ["CREATED"]
	json.dump(c, open(os.environ["CONFIG"], "w"), indent=2)
	import json, os, tempfile
	config_path = os.environ["CONFIG"]
	try:
	c = json.load(open(config_path))
	except Exception:
	c = {}
	c["name"] = os.environ["MY_NAME"]
	c["host"] = os.environ["MY_HOST"]
	c["host_target"] = os.environ["SSH_TARGET"]
	c["created"] = os.environ["CREATED"]
	config_dir = os.path.dirname(config_path) or "."
	fd, tmp_path = tempfile.mkstemp(dir=config_dir, prefix=".config.", suffix=".tmp")
	try:
	with os.fdopen(fd, "w") as f:
	json.dump(c, f, indent=2)
	f.flush()
	os.fsync(f.fileno())
	os.replace(tmp_path, config_path)
	finally:
	if os.path.exists(tmp_path):
	os.unlink(tmp_path)

[Copilot]

Same atomicity concern as the joiner branch: this merges into config.json with an in-place write. Since other subprocesses (monitor/identity/whois) can read config.json while airc connect is running, a non-atomic write risks corrupted/empty reads. Prefer an atomic write pattern (temp file + replace) to make config updates crash-safe and race-safe.