fix(infra): bundle skills/ in wheel, surface install failures, rename /bicameral:* → /bicameral-*

.claude/settings.json

@@ -25,7 +25,7 @@
         "hooks": [
           {
             "type": "command",
-            "command": "[ -d .bicameral ] && [ -z \"$BICAMERAL_SESSION_END_RUNNING\" ] && BICAMERAL_SESSION_END_RUNNING=1 claude -p '/bicameral:capture-corrections --auto-ingest' || true"
+            "command": "[ -d .bicameral ] && [ -z \"$BICAMERAL_SESSION_END_RUNNING\" ] && BICAMERAL_SESSION_END_RUNNING=1 claude -p '/bicameral-capture-corrections --auto-ingest' || true"
           }
         ]
       }

README.md

@@ -195,7 +195,7 @@ Running `bicameral-mcp setup` writes these files to your repo:
 | `.gitignore` entry | Ignores `.bicameral/` in solo mode | Recommended |
 | `.claude/settings.json` | PostToolUse hook: auto-calls `bicameral.link_commit` after git commits | Optional — improves sync |
 | `.claude/settings.json` | SessionEnd hook: runs `bicameral-capture-corrections` to catch uningested mid-session decisions | Optional — closes correction capture gap |
-| `.claude/skills/bicameral-*/SKILL.md` | Slash commands (`/bicameral:ingest`, `/bicameral:preflight`, `/bicameral:capture-corrections`, etc.) | Recommended |
+| `.claude/skills/bicameral-*/SKILL.md` | Slash commands (`/bicameral-ingest`, `/bicameral-preflight`, `/bicameral-capture-corrections`, etc.) | Recommended |
 
 ### Removing Bicameral
 
@@ -227,11 +227,11 @@ After setup, Claude Code gets these slash commands:
 
 | Command | When to use |
 |---|---|
-| `/bicameral:ingest` | Paste a transcript, PRD, or Slack thread to track its decisions |
-| `/bicameral:preflight` | Surface relevant decisions and drift before implementing |
-| `/bicameral:history` | See all tracked decisions grouped by feature area |
-| `/bicameral:dashboard` | Open the live decision dashboard in your browser |
-| `/bicameral:reset` | Wipe and replay the ledger (emergency use) |
+| `/bicameral-ingest` | Paste a transcript, PRD, or Slack thread to track its decisions |
+| `/bicameral-preflight` | Surface relevant decisions and drift before implementing |
+| `/bicameral-history` | See all tracked decisions grouped by feature area |
+| `/bicameral-dashboard` | Open the live decision dashboard in your browser |
+| `/bicameral-reset` | Wipe and replay the ledger (emergency use) |
 
 The agent also fires these automatically — `preflight` before any code change, `ingest` when you paste a document.
 

assets/git-for-specs-deck.html

@@ -169,7 +169,7 @@ <h2 class="t-h1 rv">What This Looks Like in Practice</h2>
       Symbol: <span style="color:var(--accent)">PaymentProcessor.charge</span><br>
       Evidence: <span style="color:var(--red)">PayPal import added</span>
     </div>
-    <div style="font-size:clamp(9px,1vw,11px);color:var(--text-dim)">Commit blocked. Resolve with <span class="chip">/bicameral:supersede</span> or <span class="chip">--no-verify</span></div>
+    <div style="font-size:clamp(9px,1vw,11px);color:var(--text-dim)">Commit blocked. Resolve with <span class="chip">/bicameral-supersede</span> or <span class="chip">--no-verify</span></div>
     <div style="font-size:clamp(9px,1vw,11px);color:var(--text-dim)">Decision context shared via <strong style="color:var(--text)">version control</strong>, allowing for organic <strong style="color:var(--text)">cross-functional context-sharing</strong> and alignment</div>
   </div>
 

docs/DEV_CYCLE.md

@@ -64,7 +64,7 @@ Observable evidence that a real user / agent / contributor stubbed their
 toe on something that should "just work." Symptoms, not fixes.
 
 Examples:
-- Slack thread from a design partner showing `claude -p '/bicameral:sync'`
+- Slack thread from a design partner showing `claude -p '/bicameral-sync'`
   exiting silently (#124).
 - Dashboard footage of a mid-session constraint orphaning as a parallel
   decision instead of linking to its parent.

pyproject.toml

@@ -50,6 +50,7 @@ test = [
   "tiktoken>=0.7.0,<1.0.0",
   "ruff>=0.5.0",
   "mypy>=1.10.0",
+  "build>=1.0.0",
 ]
 
 [project.scripts]
@@ -70,6 +71,13 @@ exclude = [
 ]
 artifacts = ["skills/**/*.md", "skills/**/*.yaml"]
 
+# `force-include` ships skill source unconditionally, regardless of Hatchling's
+# package-discovery rules. Required because `skills/` has no `__init__.py` and
+# `artifacts` alone does not bundle these files into the wheel (verified empty
+# pre-fix). Locked by tests/test_installer_packaging.py.
+[tool.hatch.build.targets.wheel.force-include]
+"skills" = "skills"
+
 [tool.ruff]
 line-length = 100
 target-version = "py311"

scripts/hooks/post_commit_sync_reminder.py

@@ -2,7 +2,7 @@
 
 When the agent runs ``git commit`` / ``git merge`` / ``git pull`` /
 ``git rebase --continue``, inject a system-reminder telling the agent to
-call ``/bicameral:sync`` so the decision ledger picks up the new HEAD,
+call ``/bicameral-sync`` so the decision ledger picks up the new HEAD,
 runs compliance checks, and produces authoritative reflected/drifted
 verdicts before the next user turn.
 
@@ -12,7 +12,7 @@
 from PostToolUse hooks is silently dropped to the debug log — only
 UserPromptSubmit / UserPromptExpansion / SessionStart treat raw stdout
 as agent-visible context. Symptom: the agent committed but never
-followed through to call ``link_commit`` / ``/bicameral:sync`` because
+followed through to call ``link_commit`` / ``/bicameral-sync`` because
 the reminder never reached the model. Fix: emit the structured
 envelope ``{"hookSpecificOutput": {"hookEventName": "PostToolUse",
 "additionalContext": "..."}}``.
@@ -33,7 +33,7 @@
 BASH_TOOL_NAME = "Bash"
 
 # Substrings that mark a git write-op against HEAD that the agent should
-# follow up with /bicameral:sync. Exact phrasing matches the legacy
+# follow up with /bicameral-sync. Exact phrasing matches the legacy
 # inline command's tuple so behavior is byte-identical except for the
 # stdout envelope.
 WRITE_OP_MARKERS: tuple[str, ...] = (
@@ -44,7 +44,7 @@
 )
 
 REMINDER_TEXT = (
-    "bicameral: new commit detected — run /bicameral:sync to resolve "
+    "bicameral: new commit detected — run /bicameral-sync to resolve "
     "compliance and get authoritative reflected/drifted status"
 )
 

server.py

@@ -133,7 +133,7 @@ async def list_tools() -> list[Tool]:
                 "Sync a commit into the decision ledger. Updates implemented_by/touches edges, "
                 "recomputes content hashes, re-evaluates drift for affected decisions. "
                 "Idempotent — calling twice for the same commit is a no-op. "
-                "Slash alias: /bicameral:link-commit"
+                "Slash alias: /bicameral-link-commit"
             ),
             inputSchema={
                 "type": "object",
@@ -169,7 +169,7 @@ async def list_tools() -> list[Tool]:
                 "At least one text field per decision must be non-empty or the decision is silently dropped. "
                 "The `query` field drives the post-ingest auto-brief and gap-judge chain — always pass it. "
                 "Auto-grounds decisions to code via semantic search over the symbol graph. Ensures the code index is fresh before grounding. "
-                "Slash alias: /bicameral:ingest"
+                "Slash alias: /bicameral-ingest"
             ),
             inputSchema={
                 "type": "object",
@@ -202,7 +202,7 @@ async def list_tools() -> list[Tool]:
                 "Pass start_line/end_line when you have exact lines (e.g. from a Read call) — "
                 "omit them to let the server resolve the exact line range automatically. Binding the same "
                 "(decision, region) pair twice is idempotent. "
-                "Slash alias: /bicameral:bind"
+                "Slash alias: /bicameral-bind"
             ),
             inputSchema={
                 "type": "object",
@@ -288,7 +288,7 @@ async def list_tools() -> list[Tool]:
                 "before confirming full mode. "
                 "DRY RUN BY DEFAULT — confirm=false returns the wipe plan without touching anything. "
                 "Pass confirm=true to actually wipe. "
-                "Slash alias: /bicameral:reset"
+                "Slash alias: /bicameral-reset"
             ),
             inputSchema={
                 "type": "object",
@@ -332,7 +332,7 @@ async def list_tools() -> list[Tool]:
                 "When fired=false, the agent MUST produce no output and proceed silently — "
                 "that's the trust contract. When fired=true, render the surfaced context with "
                 "a '(bicameral surfaced)' attribution before continuing with the implementation. "
-                "Slash alias: /bicameral:preflight"
+                "Slash alias: /bicameral-preflight"
             ),
             inputSchema={
                 "type": "object",
@@ -389,7 +389,7 @@ async def list_tools() -> list[Tool]:
                 "standalone. Rubric categories: missing_acceptance_criteria, "
                 "underdefined_edge_cases, infrastructure_gap, underspecified_integration, "
                 "missing_data_requirements. "
-                "Slash alias: /bicameral:judge_gaps"
+                "Slash alias: /bicameral-judge_gaps"
             ),
             inputSchema={
                 "type": "object",
@@ -427,7 +427,7 @@ async def list_tools() -> list[Tool]:
                 "decision/region IDs are returned as structured rejections (not "
                 "exceptions) so the caller can retry the accepted subset. The server "
                 "never calls an LLM — every semantic judgment lives in the caller's "
-                "session. Slash alias: /bicameral:resolve_compliance"
+                "session. Slash alias: /bicameral-resolve_compliance"
             ),
             inputSchema={
                 "type": "object",
@@ -506,7 +506,7 @@ async def list_tools() -> list[Tool]:
                 "as a negative signal; agents consult it to avoid implementing what the team rejected. "
                 "Both actions are idempotent (was_new=false if already in that state). "
                 "The signer field identifies the human or agent; the optional note captures the rationale. "
-                "Slash alias: /bicameral:ratify"
+                "Slash alias: /bicameral-ratify"
             ),
             inputSchema={
                 "type": "object",
@@ -554,7 +554,7 @@ async def list_tools() -> list[Tool]:
                 "Writes an input_span→context_for→decision edge (confirmed or rejected). "
                 "Context-pending decisions with ≥1 confirmed context_for edge become eligible "
                 "for bicameral.ratify. "
-                "Slash alias: /bicameral:resolve-collision"
+                "Slash alias: /bicameral-resolve-collision"
             ),
             inputSchema={
                 "type": "object",
@@ -597,7 +597,7 @@ async def list_tools() -> list[Tool]:
                 "Capped at 50 features; use feature_filter to drill in when truncated=True. "
                 "Does NOT fire on implementation, ingest, or drift-specific queries — use "
                 "bicameral.preflight or bicameral.ingest for those. "
-                "Slash alias: /bicameral:history"
+                "Slash alias: /bicameral-history"
             ),
             inputSchema={
                 "type": "object",
@@ -629,7 +629,7 @@ async def list_tools() -> list[Tool]:
                 "Subsequent calls return the existing URL immediately — the server "
                 "is a singleton and stays running for the session. "
                 "Fires on: 'open dashboard', 'show live history', 'launch dashboard'. "
-                "Slash alias: /bicameral:dashboard"
+                "Slash alias: /bicameral-dashboard"
             ),
             inputSchema={
                 "type": "object",

setup_wizard.py

@@ -165,22 +165,30 @@ def _select_agents() -> list[str]:
     return selected
 
 
+class RunnerNotFoundError(RuntimeError):
+    """Raised when no viable runner for bicameral-mcp is on PATH.
+
+    There is no `bicameral_mcp` package — the entry point is `server:cli_main` —
+    so `python -m bicameral_mcp` is not a valid fallback.
+    """
+
+
 def _detect_runner() -> tuple[str, list[str]]:
     """Detect the best available runner for bicameral-mcp.
 
-    Preference order:
-      1. bicameral-mcp binary on PATH — uses the actual installed environment,
-         so local subpackages (dashboard/, etc.) and editable installs work.
-      2. python3 -m bicameral_mcp — fallback for source checkouts / venvs.
+    Only one runner is supported: the `bicameral-mcp` script installed by
+    `pip install bicameral-mcp` (or an editable install in a venv). pipx run
+    is intentionally NOT used: it downloads a fresh ephemeral copy from PyPI
+    on every server start, missing local-only modules and risking version skew.
 
-    pipx run is intentionally NOT used: it downloads a fresh ephemeral copy
-    from PyPI on every server start, which misses local-only modules and can
-    run a different version than what the user installed.
+    The previous `python -m bicameral_mcp` fallback was removed because there
+    is no `bicameral_mcp` package; the resulting MCP config was non-functional.
     """
     if shutil.which("bicameral-mcp"):
         return ("bicameral-mcp", [])
-    python = "python3" if shutil.which("python3") else "python"
-    return (python, ["-m", "bicameral_mcp"])
+    raise RunnerNotFoundError(
+        "No runner found for bicameral-mcp. Install with: pip install bicameral-mcp"
+    )
 
 
 def _build_config(
@@ -384,7 +392,7 @@ def _build_session_end_command(mcp_config_path: str | None = None) -> str:
     return (
         '[ -d .bicameral ] && [ -z "$BICAMERAL_SESSION_END_RUNNING" ] && '
         "BICAMERAL_SESSION_END_RUNNING=1 "
-        f"claude -p '/bicameral:capture-corrections --auto-ingest'{extra_flags} || true"
+        f"claude -p '/bicameral-capture-corrections --auto-ingest'{extra_flags} || true"
     )
 
 
@@ -396,7 +404,7 @@ def _build_session_end_command(mcp_config_path: str | None = None) -> str:
 # Fires after every Bash tool use. When the command is a git write-op
 # (commit / merge / pull / rebase --continue), emits a hookSpecificOutput
 # envelope whose additionalContext nudges the agent to invoke
-# /bicameral:sync — running the full link_commit → compliance check
+# /bicameral-sync — running the full link_commit → compliance check
 # flow so status is authoritative immediately.
 #
 # Was a plain-stdout python -c one-liner. Per Claude Code 2.x hook docs
@@ -630,6 +638,9 @@ def _install_skills(repo_path: Path) -> int:
     """Copy skill definitions into .claude/skills/ in the target repo."""
     skills_src = Path(__file__).parent / "skills"
     if not skills_src.exists():
+        print(f"  WARNING: skill source not found at {skills_src}")
+        print("  Skills were not installed. The wheel may have been built without skills/.")
+        print("  Re-install bicameral-mcp from a recent release, or report this bug.")
         return 0
 
     skills_dst = repo_path / ".claude" / "skills"
@@ -867,11 +878,11 @@ def run_setup(
     agents = _select_agents()
 
     # Step 3: Runner check
-    command, _ = _detect_runner()
-    if command not in ("bicameral-mcp",):
-        print("\n  Note: bicameral-mcp binary not found on PATH.")
-        print(f"  Using '{command} -m bicameral_mcp' as runner.")
-        print("  Install for a cleaner setup: pip install bicameral-mcp")
+    try:
+        _detect_runner()
+    except RunnerNotFoundError as e:
+        print(f"\n  ERROR: {e}")
+        return 1
 
     # Step 4: Collaboration mode + guided intensity + telemetry + gitignore
     collab_mode = _select_collaboration_mode()
@@ -894,8 +905,7 @@ def run_setup(
     # Step 6: Install skills + hooks (Claude Code only)
     if "claude" in agents:
         num_skills = _install_skills(repo_path)
-        if num_skills:
-            print(f"  Claude Code: installed {num_skills} slash commands")
+        print(f"  Claude Code: installed {num_skills} skill(s) at {repo_path}/.claude/skills/")
         if _install_claude_hooks(repo_path):
             print(
                 "  Claude Code: installed hooks → link_commit on commit · capture-corrections on session end"
@@ -927,11 +937,11 @@ def run_setup(
 
     if "claude" in agents:
         print("  Claude Code slash commands:")
-        print("    /bicameral:ingest     — ingest a transcript, Slack thread, or PRD")
-        print("    /bicameral:preflight  — pre-flight: surface decisions before coding")
-        print("    /bicameral:history    — list all tracked decisions by feature area")
-        print("    /bicameral:dashboard  — open live decision dashboard in browser")
-        print("    /bicameral:reset      — nuke and replay the ledger (emergency)")
+        print("    /bicameral-ingest     — ingest a transcript, Slack thread, or PRD")
+        print("    /bicameral-preflight  — pre-flight: surface decisions before coding")
+        print("    /bicameral-history    — list all tracked decisions by feature area")
+        print("    /bicameral-dashboard  — open live decision dashboard in browser")
+        print("    /bicameral-reset      — nuke and replay the ledger (emergency)")
         print()
 
     print("  Or just ask naturally:")

skills/bicameral-capture-corrections/SKILL.md

@@ -146,7 +146,7 @@ capture-corrections output.
 ## SessionEnd batch mode
 
 Fires via the `SessionEnd` hook in `.claude/settings.json`. Also invocable
-manually as `/bicameral:capture-corrections`.
+manually as `/bicameral-capture-corrections`.
 
 ### Steps
 
@@ -235,7 +235,7 @@ user's project `.claude/settings.json`. No manual configuration needed.
 
 Command written by the setup wizard:
 ```
-[ -d .bicameral ] && [ -z "$BICAMERAL_SESSION_END_RUNNING" ] && BICAMERAL_SESSION_END_RUNNING=1 claude -p '/bicameral:capture-corrections --auto-ingest' || true
+[ -d .bicameral ] && [ -z "$BICAMERAL_SESSION_END_RUNNING" ] && BICAMERAL_SESSION_END_RUNNING=1 claude -p '/bicameral-capture-corrections --auto-ingest' || true
 ```
 
 Two guards:

skills/bicameral-config/SKILL.md

@@ -1,6 +1,6 @@
-# /bicameral:config — Interactive Configuration
+# /bicameral-config — Interactive Configuration
 
-**Trigger**: user types `/bicameral:config`
+**Trigger**: user types `/bicameral-config`
 
 Walk through each bicameral configuration setting interactively, write the
 updated `config.yaml`, and reinstall all hooks so changes take effect

skills/bicameral-preflight/SKILL.md

@@ -244,7 +244,7 @@ so you can see what your branch changes relative to main.
 ### 3.5 Scan recent user turns for uningested corrections
 
 Before classifying server-returned findings, invoke
-`/bicameral:capture-corrections` in **in-session mode**:
+`/bicameral-capture-corrections` in **in-session mode**:
 
 ```
 Skill("bicameral:capture-corrections", args="--mode in-session")

skills/bicameral-reset/SKILL.md

@@ -25,7 +25,7 @@ The fail-safe valve. When the ledger gets polluted — bad ingest, stale groundi
 ## When NOT to fire
 
 - **Never fire automatically.** Reset is always user-initiated.
-- Drift reports that look wrong → run `/bicameral:sync` first, escalate to reset only if that doesn't help.
+- Drift reports that look wrong → run `/bicameral-sync` first, escalate to reset only if that doesn't help.
 - If only one ingest looks bad, suggest re-running that ingest rather than wiping everything.
 
 ## The two-call pattern (always)