yugabyte/meko-skills/meko-bmad-context

meko-bmad-context — Recall Meko context for a BMad workflow (Claude Code)

Retrieve the small set of prior experience that is actually relevant to the BMad workflow now running, so the workflow adapts using past decisions, evidence, and failures — without loading whole documents into the prompt.

Inputs

• workflow-name (required) — the BMad workflow invoking recall, e.g. bmad-prd.
• objective (required) — what this workflow run is trying to produce.
• focused-query (optional) — a narrower topic to bias retrieval.

Identifiers — read, never invent

Read from the SessionStart-injected Meko context:

• agent_id — use it verbatim.
• conversation_id — use it verbatim for trace nesting.
• active datapack_id — pass to every datapack-scoped call.

If any required identifier is absent, do not fabricate it. Skip the calls that need it, warn once, and let the BMad workflow continue normally.

Procedure

1. Memory recall. Call memory_search with a focused query built from the

objective (and focused-query if given) covering: prior decisions, constraints, preferences, unresolved risks, and related BMad outputs. Pass agent_id, conversation_id, and datapack_id. Bound to 5 results.

1. Shared knowledge. Only when an active datapack_id exists and shared

project or organizational knowledge is relevant, call knowledgebase_search with datapack_id (required). Bound to 5 results. Skip this call entirely when there is no datapack id.

1. Synthesize. Return the compact structure below. Keep it short — recall

points to relevant experience; it does not dump documents.

1. Authority. Current user instructions and repository state are

authoritative. When recalled context conflicts with them, surface the conflict and defer to current evidence.

1. Graceful degradation. If Meko is unavailable or returns nothing useful,

emit a single warning and let the workflow proceed with its normal local behavior. Recall is never a completion dependency.

Return structure

Return this shape (omit empty sections, keep each list tight):

recalled_facts:      # up to 5 — prior decisions / constraints / preferences / risks, each with a one-line why
shared_knowledge:    # up to 5 — relevant datapack knowledge snippets (only if datapack present)
provenance:          # ids/hashes/sources backing the above (memory ids, artifact hashes)
conflicts:           # recalled items that contradict current instructions/repo — flagged, deferring to current
warnings:            # e.g. "Meko unavailable", "no datapack id", "no relevant context found"

See references/retrieval.md for query-construction guidance and worked examples.