Enter Old Major session mode — fetch personas, assemble session stack, activate
npx skills add https://github.com/MockaSort-Studio/hall-of-automata-cli --skill hall-open复制此命令并粘贴到 Claude Code 中以安装该技能
Dispatch ready tasks to the Hall as GitHub Issues with quota stewardship
Dump the current plan as JSON, Markdown, and/or Mermaid diagram
Resync local plan state from GitHub issue/PR states
Provision the Hall of Automata Projects v2 board, custom fields, and labels on the current target repo — idempotent
List, view, or prune saved Tier-2 subagent consultation outputs
Clean up old plan directories or stale persona cache from .hall-cache/
| name | hall-open |
| description | Enter Old Major session mode — fetch personas, assemble session stack, activate |
| argument-hint | ["--refresh|--verify"] |
| allowed-tools | ["Bash","Write","AskUserQuestion","CronCreate","mcp__github__get_file_contents","mcp__github__get_me","mcp__github__get_team_members","mcp__github__search_repositories","mcp__hall-projects__read_board"] |
Enter Hall session mode. Fetches personas, assembles session stack, activates Old Major.
Use --refresh to force persona re-fetch even if cache is fresh. Use --verify to force invoker re-check.
Execute each step in order. Hard-stop on any error; warn-and-continue on non-critical issues.
Flag pre-processing:
--verify was passed: rm -f .hall-cache/invoker.json--refresh was passed: treat NEED_FETCH=true regardless of the block output below.set -euo pipefail
# Hard stops
gh auth status &>/dev/null || { echo "ERROR: gh not authenticated" >&2; exit 1; }
ORIGIN=$(git remote get-url origin 2>/dev/null || echo "")
STANDALONE=$([ -z "$ORIGIN" ] && echo true || echo false)
if [ "$STANDALONE" = "false" ]; then
REPO=$(echo "$ORIGIN" | sed 's|.*github.com[:/]||;s|\.git$||')
fi
[ -n "${GITHUB_PERSONAL_ACCESS_TOKEN:-}" ] || echo "WARN: GITHUB_PERSONAL_ACCESS_TOKEN not set — MCP unavailable."
# Gitignore
grep -q "\.hall-cache" .gitignore 2>/dev/null \
|| { echo ".hall-cache/" >> .gitignore; echo "Added .hall-cache/ to .gitignore"; }
# Cache state
mkdir -p .hall-cache/personas .hall-cache/session .hall-cache/plans
Call get_file_contents MCP: owner=MockaSort-Studio, repo=hall-of-automata, path=agents.yml. Extract sha → CURRENT_SHA.
# On rate_limit/secondary-rate-limit error: gh api repos/MockaSort-Studio/hall-of-automata/contents/agents.yml --jq '.sha'
CACHED_SHA=$(cat .hall-cache/personas/.agents-yml-sha 2>/dev/null || echo "")
FETCHED_AT=$(cat .hall-cache/personas/.fetched_at 2>/dev/null || echo "")
NOW=$(date +%s)
FETCHED_TS=$([ -n "$FETCHED_AT" ] && date -d "$FETCHED_AT" +%s 2>/dev/null || echo "0")
NEED_FETCH=false
[ "$CURRENT_SHA" != "$CACHED_SHA" ] && NEED_FETCH=true
[ $(( NOW - FETCHED_TS )) -gt 86400 ] && NEED_FETCH=true
[ -z "$FETCHED_AT" ] && NEED_FETCH=true
python3 -c "import json; d=json.load(open('.hall-cache/personas/.advisory-roster.json')); assert isinstance(d,list)" 2>/dev/null \
|| NEED_FETCH=true
ACTIVE_PLAN=false
for d in .hall-cache/plans/*/; do
f="${d}plan.md"; [ -f "$f" ] || continue
grep -qm1 "Status:.*DONE" "$f" 2>/dev/null || { ACTIVE_PLAN=true; break; }
done
AUTO_LEVEL=$(python3 -c "import json; print(json.load(open('.hall-cache/session/config.json')).get('automation_level','missing'))" \
2>/dev/null || echo "missing")
LOCAL_MODE=$(python3 -c "import json; print(json.load(open('.hall-cache/session/config.json')).get('local_mode','missing'))" \
2>/dev/null || echo "missing")
echo "$CURRENT_SHA" > .hall-cache/session/.current-sha
echo "STANDALONE=$STANDALONE | NEED_FETCH=$NEED_FETCH | ACTIVE_PLAN=$ACTIVE_PLAN | AUTO_LEVEL=$AUTO_LEVEL | LOCAL_MODE=$LOCAL_MODE"
echo "CONTEXT_EXISTS=$([ -f .hall-cache/session/context.md ] && echo true || echo false)"
echo "SHA=${CURRENT_SHA:0:8}"
If STANDALONE=true: read skills/hall-open/standalone-flow.md (resolve against $CLAUDE_PLUGIN_ROOT) and execute the org/repo resolution procedure exactly as specified. On completion, ORG, REPO_NAME, and REPO are set for subsequent steps.
Read CURRENT_SHA from .hall-cache/session/.current-sha; if absent, call get_file_contents MCP (owner=MockaSort-Studio, repo=hall-of-automata, path=agents.yml) and extract sha.
# On rate_limit/secondary-rate-limit error: gh api repos/MockaSort-Studio/hall-of-automata/contents/agents.yml --jq '.sha'
Call get_file_contents MCP: owner=MockaSort-Studio, repo=hall-of-automata, path=roster/. From the returned array, keep entries where type=file, name ends in .md, name ≠ old-major.md and ≠ README.md. Write the names (without .md) as a JSON array to .hall-cache/personas/.advisory-roster.json.
# On rate_limit/secondary-rate-limit error: gh api repos/MockaSort-Studio/hall-of-automata/contents/roster --jq '[.[] | select(.type=="file" and (.name|endswith(".md")) and .name!="old-major.md" and .name!="README.md") | .name[:-3]]' > .hall-cache/personas/.advisory-roster.json
python3 -c "import json,sys; d=json.load(open('.hall-cache/personas/.advisory-roster.json')); assert isinstance(d,list), f'API error: {d}'" \
|| exit 1
Fetch and write persona files via Bash (the Write tool fails on new files; gh api writes directly):
SPECS=$(python3 -c "import json; print(' '.join(json.load(open('.hall-cache/personas/.advisory-roster.json'))))")
gh api repos/MockaSort-Studio/hall-of-automata/contents/agents/automaton_base.md \
--jq '.content' | base64 -d > .hall-cache/personas/automaton_base.md
gh api repos/MockaSort-Studio/hall-of-automata/contents/roster/old-major.md \
--jq '.content' | base64 -d > .hall-cache/personas/old-major.md
for name in $SPECS; do
gh api "repos/MockaSort-Studio/hall-of-automata/contents/roster/${name}.md" \
--jq '.content' | base64 -d > ".hall-cache/personas/${name}.md"
done
CURRENT_SHA="$CURRENT_SHA" python3 "$CLAUDE_PLUGIN_ROOT/scripts/verify-personas.py"
--refresh limitation: Stack changes regenerated in --refresh don't take effect in the current context window — the @-import chain is evaluated only at conversation start. A fresh cc session is required for persona or methodology changes to apply. See Step 5.
Read skills/hall-open/session-setup.md (resolve against $CLAUDE_PLUGIN_ROOT) and execute the session setup procedure exactly as specified.
Read the first 30 lines of README.md and write a 2–4 sentence brief to .hall-cache/session/context.md. If no README: Project context: not available.
Standalone mode: if STANDALONE=true, call get_file_contents MCP (owner=$ORG, repo=$REPO_NAME, path=CLAUDE.md). On success, write decoded content to ~/.hall/context/target-claude.md; incorporate as supplemental project context in context.md. On 404: skip silently; synthesise from README only.
Read .hall-cache/session/CLAUDE-stack.md and each @-imported file in order; apply as operating instructions. Skip if resume mode and --refresh was not passed — stack already loaded via CLAUDE.md @-imports. On --refresh: always run this step regardless of mode; @-import chains are not re-evaluated mid-session, so the explicit read makes regenerated stack content active immediately.
Read skills/hall-open/invoker-gate.md (resolve against $CLAUDE_PLUGIN_ROOT) and execute the invoker detection procedure exactly as specified.
ls .hall-cache/plans/ 2>/dev/null || true
List existing plans with status. Ask whether to resume or start fresh. Then ask what the invoker wants to build — one sentence, in character as Old Major.