@tekmidian/pai 0.40.0 → 0.41.0

package/dist/skills/Worker/SKILL.md

@@ -25,12 +25,22 @@ pai worker run --label "short task label" --class research \\
   `pai worker run --chain draft,implement …` — the draft stage turns the brief
   into a spec file, implement runs with it, `--chain draft,implement,review`
   adds a review pass. `--class spotcheck` for verification runs.
+- Grant MCP tools by naming mcp__server__tool in --allowedTools (the server loads automatically).
 - The answer is in the `result` field of the JSON it prints. Review the diff yourself.
 - `--no-pane` suppresses the iTerm follow pane; panes open automatically otherwise.
 - `--agent <name>` runs a definition from ~/.claude/agents/<name>.md (the agent
   library runs on workers: body becomes the system prompt, tools the allowlist,
   model the class).
 
+### Waiting
+
+- NEVER busy-wait for a worker. No sleep loops, no `sleep N; pai worker ps`
+  polling, no manual retry loops. Two sanctioned waits: run workers as
+  background Bash tasks (the harness fires a completion notification), or
+  call `pai worker wait <id...>` which blocks until they finish and prints
+  each result as one JSON line. If you catch yourself sleeping to re-check a
+  worker, stop — you already get notified.
+
 ### Watching
 
 - `worker_ps` — running + last finished workers (chains show as trees).

package/docs/commands/README.md

@@ -68,8 +68,8 @@ Generated man pages for every `pai` command area. Read any page in the terminal
 | `pai kg query` | Query KG triples by subject, predicate, object, time, or project |
 | `pai kg stats` | Show triple counts and contradiction count |
 | `pai mcp` | MCP server management: install and status |
-| `pai mcp install` | Register the PAI MCP server in ~/.claude.json (restart Claude Code to activate) |
-| `pai mcp status` | Show whether the PAI MCP server is registered and the binary exists |
+| `pai mcp install` | Register the PAI MCP server (pai) in ~/.claude.json (restart Claude Code to activate) |
+| `pai mcp status` | Show whether the PAI MCP server (pai) is registered and the binary exists |
 | `pai memory` | Memory engine: index, search, and status |
 | `pai memory embed [project-slug]` | Generate embeddings for un-embedded chunks (Phase 2.5) |
 | `pai memory index [project-slug]` | Index memory files for one project or all projects |

package/docs/commands/mcp.md

@@ -14,17 +14,17 @@ pai mcp <subcommand> [options]
 
 | Command | Description |
 |---------|-------------|
-| [`pai mcp install`](#pai-mcp-install) | Register the PAI MCP server in ~/.claude.json (restart Claude Code to activate) |
-| [`pai mcp status`](#pai-mcp-status) | Show whether the PAI MCP server is registered and the binary exists |
+| [`pai mcp install`](#pai-mcp-install) | Register the PAI MCP server (pai) in ~/.claude.json (restart Claude Code to activate) |
+| [`pai mcp status`](#pai-mcp-status) | Show whether the PAI MCP server (pai) is registered and the binary exists |
 
 ### pai mcp install
 
-Register the PAI MCP server in ~/.claude.json (restart Claude Code to activate)
+Register the PAI MCP server (pai) in ~/.claude.json (restart Claude Code to activate)
 
 
 ### pai mcp status
 
-Show whether the PAI MCP server is registered and the binary exists
+Show whether the PAI MCP server (pai) is registered and the binary exists
 
 
 ## See also

package/docs/commands/worker.md

@@ -47,6 +47,7 @@ Run one claude-code worker through the configured provider.
 
 Unknown options are passed to claude verbatim (e.g. -p, --allowedTools);
 --output-format/--verbose are handled here.
+Grant MCP tools by naming mcp__server__tool in --allowedTools (the server loads automatically);
 --chain draft,implement[,review] runs a spec-first pipeline;
 --agent <name> runs an agent definition from ~/.claude/agents.
 
@@ -192,6 +193,8 @@ Merge a worker's worktree branch (worker/<id>) into the original checkout, then
 
 Poll workers until they finish; prints each result as one JSON line, exit 1 on failure or timeout
 
+Never busy-wait for a worker: no sleep loops, no sleep-then-`pai worker ps` polling, no manual retry loops. Two sanctioned waits: run workers as background Bash tasks (the harness notifies on completion), or call `pai worker wait`, which blocks until they finish and prints each result. If you catch yourself sleeping to re-check a worker, stop — you already get notified.
+
 **Arguments**
 
 | Argument | Kind |

package/docs/provider-independence.md

@@ -1,22 +1,20 @@
 # Provider Independence
 
-How to free yourself from Claude — so a maxed-out subscription stops being your problem. How to run any cloud model and any local model inside your Claude Code (or OpenCode) harness. How to run models in parallel, each doing what it is best at — all of it purely local, no need for yet another Hermes-or-whatever harness. How to have workers run by default, visibly, each with its own chat you can talk to.
+I started my Thursday at 5am, which is normal — and at about 80%, which is not.
 
-My journey into PAI and AIBroker started nine months ago, when OpenClawd burned through all my credits in four hours. I decided then and there that there had to be a better way: instead of buying API tokens, use the Claude Max subscription I was already paying for.
+Because those 80% are not motivation (that was certainly higher). That is the weekly usage of my Claude Max 200 subscription.
 
-Nine months later, I hit the next wall: maxing out Claude Max itself. The next level of independence was required. I built it — in one day. Here is what happened.
+With still two days to go, I was getting worried. Last week I had already hit 100% hard — despite all my efficiency gains, including creating my own "Agentish" language.
 
-I started my Thursday at 5am, which is normal, and at about 80%, which is not.
+At 9am I was above 90%, and there was no way I would make it through to Saturday 8am, the weekly reset.
 
-That 80% is not motivation — although that was certainly higher. It is the weekly usage of my Claude Max 200 subscription.
+So I decided to solve this problem. At 10pm I am at 97% — and that number has not changed since noon.
 
-With two days to go until the Saturday 8am reset, I was getting worried. Last week I had already hit 100% — despite all my efficiency gains, including inventing my own "Agentish" language.
+This page shows how to become provider-independent. How to free yourself from Claude, so a maxed-out subscription stops being your problem. How to use any cloud model and any local model inside your Claude Code (or OpenCode) harness. How to run models in parallel, each for what it is good at — all purely local, without resorting to yet another Hermes-or-whatever harness. How to have workers run by default, visibly — and how to interact with each of them in its own chat.
 
-By 9am I was above 90%. There was no way I would make it to the reset.
+My whole journey into PAI and AIBroker started nine months ago, when I used OpenClawd for four hours and had burned through all my credits. I decided then and there that there had to be a better way: instead of API tokens, use the Claude Max subscription I was paying for anyway.
 
-So I decided to solve this problem instead of waiting it out. By 10pm I was at 97% — a number that has not moved since noon, because since noon, almost nothing has run on Claude.
-
-This is how I did it.
+Nine months later, I ran into the next wall: maxing out the Claude Max subscription itself. So the next level of independence was required — and I made that happen, today. Here is what happened.
 
 I made myself independent of Claude and Fable. They are great — but I hate being locked in, and I suspect most of us do. I want to decide which model I use: Claude, GLM, Grok, whatever. I need to abstract away from that choice.
 
@@ -35,18 +33,7 @@ So the requirements were:
 
 The whole implementation cost 7 of my last 10 percentage points of Claude Max 200. Then I switched to GLM for everything else. And what can I say — it works beautifully.
 
----
-
-The rest of this page is the operator's guide: switching providers, starting sessions, and the three copy-paste steps to your own setup.
-
-> My AI budget hit 91% before lunch. So I changed one rule: the expensive
-> assistant keeps the thinking, a cheaper crew does the building. The feature
-> shipped anyway.
->
-> Your AI bill is not a fact of nature. It is a design decision.
 
-Your assistant runs its helper crew on whichever AI provider you choose —
-and so can the session itself.
 
 ## Switching
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tekmidian/pai",
-  "version": "0.40.0",
+  "version": "0.41.0",
   "description": "PAI Knowledge OS — Personal AI Infrastructure with federated memory and project management",
   "type": "module",
   "main": "dist/index.mjs",

package/statusline-command.sh

@@ -11,7 +11,7 @@
 #
 # LINES DISPLAYED:
 #   1. Greeting: DA name, model, directory
-#   2. MCPs: Active MCP servers (wraps on narrow terminals)
+#   2. MCPs: server count, then names that fit the width (rest shown as +N)
 #   3. Context: Current session context window usage (K / 200K)
 #
 # ENVIRONMENT VARIABLES (set in settings.json env section):
@@ -86,27 +86,26 @@ fi
 # Config directory
 claude_dir="${PAI_DIR:-$HOME/.claude}"
 
-# Count MCPs from all config sources (settings.json, .mcp.json, ~/.claude.json)
-mcp_names_raw=""
-mcps_count=0
+# Collect MCP server names from all config sources (settings.json, .mcp.json,
+# ~/.claude.json). Only entries that define a server (object with a command,
+# type or url field) count: Claude Code also stores per-tool usage stats in
+# mcpServers under tool-name keys ("Read", "mcp__pai__memory_search", …), and
+# those must never reach the status line. Keys are newline-separated so names
+# containing spaces survive.
+mcp_names=""
 
-# Helper: merge MCP names from a jq-compatible JSON file
 _merge_mcps() {
     local file="$1"
     [ -f "$file" ] || return
-    local data
-    data=$(jq -r '.mcpServers | keys | join(" "), length' "$file" 2>/dev/null)
-    [ -n "$data" ] && [ "$data" != "null" ] || return
-    local names count
-    names=$(echo "$data" | head -1)
-    count=$(echo "$data" | tail -1)
+    local names
+    names=$(jq -r '.mcpServers // {} | to_entries[] | select(((.value | type) == "object") and (.value | (has("command") or has("type") or has("url")))) | .key' "$file" 2>/dev/null)
     [ -n "$names" ] || return
-    if [ -n "$mcp_names_raw" ]; then
-        mcp_names_raw="$mcp_names_raw $names"
+    if [ -n "$mcp_names" ]; then
+        mcp_names="${mcp_names}
+${names}"
     else
-        mcp_names_raw="$names"
+        mcp_names="$names"
     fi
-    mcps_count=$((mcps_count + count))
 }
 
 # Read from all three MCP config locations
@@ -114,11 +113,12 @@ _merge_mcps "$claude_dir/settings.json"    # legacy
 _merge_mcps "$claude_dir/.mcp.json"        # project-level
 _merge_mcps "$HOME/.claude.json"           # user-level (e.g. Coogle, DEVONthink)
 
-# Deduplicate MCP names (preserving order)
-if [ -n "$mcp_names_raw" ]; then
-    mcp_names_raw=$(echo "$mcp_names_raw" | tr ' ' '\n' | awk '!seen[$0]++' | tr '\n' ' ' | sed 's/ $//')
-    mcps_count=$(echo "$mcp_names_raw" | wc -w | tr -d ' ')
-fi
+# Deduplicate (case-insensitive — macOS config keys drift in case), first
+# spelling wins, order preserved
+mcp_list=()
+while IFS= read -r mcp_line; do
+    mcp_list+=("$mcp_line")
+done < <(printf '%s\n' "$mcp_names" | awk 'NF && !seen[tolower($0)]++')
 
 # Extract context window usage from Claude Code's JSON input (no JSONL parsing needed)
 context_pct=$(echo "$input" | jq -r '.context_window.used_percentage // 0' 2>/dev/null)
@@ -247,8 +247,7 @@ if [ "${PAI_SIMPLE_COLORS:-0}" = "1" ]; then
     MCP_DEFAULT='\033[34m'
 fi
 
-# Format MCP names with terminal-width-aware wrapping
-# Debug: log available width info (remove after testing)
+# Format the MCP segment (see the render block below)
 # Terminal width for line truncation
 # Claude Code's statusline subprocess can't detect resize (stty returns stale values).
 # To set your width:  echo 105 > ~/.claude/.statusline_width
@@ -258,8 +257,7 @@ term_width=80
 [ "$term_width" -gt 0 ] 2>/dev/null || term_width=80
 mcp_prefix_width=10  # visual width of "🔌 MCPs: " (emoji=2 + space + "MCPs: " = 10)
 
-# Build MCP output — proactively split into two lines when there are many MCPs.
-# No width detection needed: if total display chars > 60, split at the midpoint.
+# Display-name mapping for known servers
 _mcp_display_name() {
     case "$1" in
         "daemon") echo "Daemon" ;;
@@ -275,7 +273,9 @@ _mcp_display_name() {
         "macos_automator") echo "macOS" ;;
         "claude_ai_Gmail") echo "Gmail" ;;
         "claude_ai_Google_Calendar") echo "GCal" ;;
-        *) local n="$1"; echo "${n^}" ;;
+        # capitalize without bash-4 ${n^} so /bin/bash 3.2 renders names too
+        *) local n="$1"
+           printf '%s%s' "$(printf '%s' "$n" | cut -c1 | tr '[:lower:]' '[:upper:]')" "$(printf '%s' "$n" | cut -c2-)" ;;
     esac
 }
 
@@ -288,57 +288,53 @@ _mcp_formatted() {
     esac
 }
 
-# Collect all display names and calculate total width
-mcp_display_names=()
-mcp_formatted_strs=()
-total_display_width=$mcp_prefix_width  # start with "🔌 MCPs: " prefix
-total_mcps=0
-
-for mcp in $mcp_names_raw; do
-    dn=$(_mcp_display_name "$mcp")
-    fm=$(_mcp_formatted "$dn")
-    mcp_display_names+=("$dn")
-    mcp_formatted_strs+=("$fm")
-    if [ $total_mcps -gt 0 ]; then
-        total_display_width=$((total_display_width + 2))  # ", "
-    fi
-    total_display_width=$((total_display_width + ${#dn}))
-    total_mcps=$((total_mcps + 1))
-done
-
-# Decide: one line or two lines?
-# If total display width > 60 chars, split at the midpoint
+# Build the MCP segment: "N: name1, name2, … +K" — total server count first,
+# then as many display names as fit the width (hard cap regardless of width),
+# the rest collapsed into "+K". Always a single line: what doesn't fit is
+# counted, never printed.
+mcp_total=${#mcp_list[@]}
 mcp_line1=""
-mcp_line2=""
 
-if [ $total_mcps -eq 0 ]; then
+if [ "$mcp_total" -eq 0 ]; then
     mcp_line1="none"
-elif [ $total_display_width -le $term_width ]; then
-    # Single line — everything fits
-    for ((i=0; i<total_mcps; i++)); do
-        if [ $i -eq 0 ]; then
-            mcp_line1="${mcp_formatted_strs[$i]}"
-        else
-            mcp_line1="${mcp_line1}${SEPARATOR_COLOR}, ${mcp_formatted_strs[$i]}"
-        fi
-    done
 else
-    # Two lines — split at midpoint
-    split_at=$(( (total_mcps + 1) / 2 ))
-    for ((i=0; i<split_at; i++)); do
-        if [ $i -eq 0 ]; then
-            mcp_line1="${mcp_formatted_strs[$i]}"
-        else
-            mcp_line1="${mcp_line1}${SEPARATOR_COLOR}, ${mcp_formatted_strs[$i]}"
+    mcp_max_names=6
+    # Width budget for the name list: terminal minus prefix, minus the leading
+    # "N: " and a reserve for the trailing " +K" overflow marker
+    mcp_budget=$(( term_width - mcp_prefix_width - ${#mcp_total} - 2 - 6 ))
+    [ "$mcp_budget" -lt 12 ] && mcp_budget=12
+    mcp_width=0
+    mcp_shown=0
+    mcp_overflow=0
+    for mcp in "${mcp_list[@]}"; do
+        dn=$(_mcp_display_name "$mcp")
+        w_add=${#dn}
+        [ "$mcp_shown" -gt 0 ] && w_add=$(( w_add + 2 ))  # ", " separator
+        if [ "$mcp_shown" -ge "$mcp_max_names" ] || [ $(( mcp_width + w_add )) -gt "$mcp_budget" ]; then
+            mcp_overflow=$(( mcp_overflow + 1 ))
+            continue
         fi
-    done
-    for ((i=split_at; i<total_mcps; i++)); do
-        if [ $i -eq $split_at ]; then
-            mcp_line2="${mcp_formatted_strs[$i]}"
+        fm=$(_mcp_formatted "$dn")
+        if [ "$mcp_shown" -eq 0 ]; then
+            mcp_line1="${mcp_total}${SEPARATOR_COLOR}: ${RESET}${fm}"
+            mcp_width=${#mcp_total}
         else
-            mcp_line2="${mcp_line2}${SEPARATOR_COLOR}, ${mcp_formatted_strs[$i]}"
+            mcp_line1="${mcp_line1}${SEPARATOR_COLOR}, ${fm}"
+            mcp_width=$(( mcp_width + 2 ))
         fi
+        mcp_width=$(( mcp_width + ${#dn} ))
+        mcp_shown=$(( mcp_shown + 1 ))
     done
+    if [ "$mcp_shown" -eq 0 ]; then
+        # Even the first name exceeds the budget — truncate that one name
+        dn=$(_mcp_display_name "${mcp_list[0]}")
+        avail=$(( mcp_budget - 1 ))
+        [ "$avail" -lt 3 ] && avail=3
+        dn="${dn:0:$((avail - 1))}…"
+        mcp_line1="${mcp_total}${SEPARATOR_COLOR}: ${RESET}$(_mcp_formatted "$dn")"
+        mcp_overflow=$(( mcp_total - 1 ))
+    fi
+    [ "$mcp_overflow" -gt 0 ] && mcp_line1="${mcp_line1}${SEPARATOR_COLOR} +${RESET}${LINE2_ACCENT}${mcp_overflow}${RESET}"
 fi
 
 # Output the statusline
@@ -356,12 +352,8 @@ else
     printf "${line1_short}\n"
 fi
 
-# LINE 2 - MCPs (with optional wrap to second line)
+# LINE 2 - MCPs (single line, capped by width; overflow shown as +N)
 printf "${LINE2_PRIMARY}${EMOJI_PLUG} MCPs${RESET}${LINE2_PRIMARY}${SEPARATOR_COLOR}: ${RESET}${mcp_line1}${RESET}\n"
-if [ -n "$mcp_line2" ]; then
-    # Continuation line — indent to align with MCP names after "🔌 MCPs: "
-    printf "${LINE2_PRIMARY}          ${RESET}${mcp_line2}${RESET}\n"
-fi
 
 
 # Usage suffix: provider-aware. glm sessions read the Z.ai plan quota