Beads

Beads

Use this as the default Beads skill.

Current doctrine is version-aware for bd 1.0.x:

• use bd bootstrap --dry-run / bd bootstrap --yes to adopt or repair existing shared repo state
• do not use bd init --force for shared-repo recovery; inspect bd init-safety instead
• use explicit bd dolt pull / bd dolt push; there is no current bd sync command
• use bd comment for concurrent observations and bd note for consolidated issue summaries
• use bd batch for scripted multi-bead mutations

The other Beads skills are implementation details:

• Bootstrap/SKILL.md for repo-local adoption and alignment
• UpgradePath/SKILL.md for tier/doctrine upgrades in repos that already use Beads
• HookAudit/SKILL.md for machine-level Claude hook repair
• VersionAudit/SKILL.md for live backend/remotes/hooks/drift inspection
• TopologyPlanner/SKILL.md for choosing the right collaboration topology
• FormulaBuilder/SKILL.md for recurring workflow templates and formulas
• EvidenceTriage/SKILL.md for evidence-first issue creation
• RecoveryAudit/SKILL.md for backup/sync/lock recovery rehearsal
• Research/SKILL.md for source-grounded doctrine and current-versus-historical guidance
• Workflows/MigrateRepo.md for applying current Beads doctrine to one repo
• Tools/AuditRepos.sh for scanning many repos before rollout

Outcome rule:

• using this skill should leave the relevant repo or machine Beads-clean when safe, not make the user remember which Beads sub-skill to invoke
• choose the smallest combination of bootstrap, upgrade, and hook repair that reaches doctrine hygiene directly
• leave manual follow-up only when blocked by permissions, credentials, or a genuinely risky coordination window
• when repo-local doctrine is in scope, make sure issue creation defaults produce self-contained beads at creation time: description for what/why, --context for execution details, and --notes for source provenance
• when a bead is clearly an epic or multi-step work package, default to immediate child decomposition with explicit dependencies rather than leaving the execution plan implied in prose
• prefer evidence-first beads over vague reminders: repro, observed behavior, likely fix surface, and acceptance criteria beat generic TODO phrasing
• treat stale epics and disproven theories as first-class cleanup work: close or supersede them explicitly instead of letting the queue rot

Routing

Start with a quick audit of the current repo and, when relevant, machine hook state.

Workflow Routing

Route to Bootstrap

Use when:

• .beads/ is missing
• repo-local AGENTS.md / CLAUDE.md are missing or clearly drifted
• the repo needs initial Beads adoption or canonical workflow alignment

Route to UpgradePath

Use when:

• the repo already uses Beads
• the main question is tier/doctrine fit rather than initial setup
• the repo likely belongs in T2 or T3 and current docs underfit that

Route to HookAudit

Use when:

• the user mentions hooks, global defaults, Claude startup behavior, or machine-level Beads behavior
• repo-local doctrine depends on healthy Claude hook behavior
• the user wants Beads made foolproof across sessions, not just inside one repo

Combined default

If the request is generic, use this order:

1. fix repo-local hygiene first
2. repair machine hooks if they are in scope or clearly undermining repo doctrine
3. apply tier/doctrine upgrades if the repo already uses Beads but is on the wrong operating model

Default scope rules:

• make Beads work in this repo => repo-local only unless machine drift is clearly in the way
• make Beads foolproof or standardize my Beads workflow => repo-local plus machine hook audit
• upgrade this repo's Beads workflow => repo-local upgrade, not machine changes, unless hook drift blocks the target behavior
• repo-local hygiene includes creation-time bead quality; do not leave repos on a workflow where bd create produces hollow tasks that need a later enrichment pass

Minimal audit

Use only what is needed to route correctly:

pwd
git rev-parse --show-toplevel
test -d .beads && echo ".beads exists" || echo ".beads missing"
bd version
bd context --json || true
bd bootstrap --dry-run || true
find . \
  -path './.git' -prune -o \
  -path './.claude/worktrees' -prune -o \
  \( -name AGENTS.md -o -name CLAUDE.md \) -print
bd ready --json || true
sed -n '1,220p' ${AGENT_HOME}/settings.json 2>/dev/null || true

Then choose the minimum justified route and do the work directly.

If repo-local doctrine is changed or aligned, verify it mechanically with:

python3 ${PAI_EXTENSIONS_DIR}/skills/Beads/Bootstrap/scripts/check_beads_doctrine.py --path .

When restructuring an epic into child beads, use this default sequence:

1. create the parent epic with full description, context, acceptance, and provenance
2. create child beads immediately when the execution steps are already clear
3. give each child bead its own description, execution context, acceptance criteria, and provenance notes
4. wire dependencies explicitly with bd dep instead of leaving order only in prose
5. leave a decomposition note on the parent listing the child IDs and intended sequence
6. run bd dolt push after the bead-tree restructure so the new plan is shared durably

Verification

Verify only the surfaces touched:

• repo-local changes: bd ready --json, git diff --check, relevant AGENTS.md / CLAUDE.md
• hook changes: bd setup claude --check, grep or sed over ${AGENT_HOME}/settings.json and referenced hook files
• shared repo setup: bd dolt remote list, bd dolt pull || true, bd dolt push

Report:

• which sub-skill logic was used
• what was changed
• what was deliberately left alone
• any residual manual step and why it could not be automated safely