/hydrate

Keep Hydra DB aligned with a fast-moving codebase. This is an incremental rewrite of one document, not a first-principles ingest and not a scatter of new source ids.

Default flow:

1. Explore codebase (delta-first via Sync waypoint when valid)
2. hydradb inspect {project}-standing --collection {project}
3. Patch sections in that document; ingest the rewritten doc as knowledge under the same id (--kind knowledge --no-infer --collection {project})
4. Advance the Sync waypoint to current HEAD

If {project}-standing is missing/empty, stop and recommend /ingest (or offer to run ingest if the user confirms).

Hard rules

• Code wins when Hydra and code disagree. Fix standing in the same run.
• Do not dump large markdown docs into context or Hydra as raw paste.
• Never store secrets, env values, tokens, or credentials.
• One project id only: {project}-standing. Never create sibling ids (*-codebase-map, *-decisions-*, *-scars, *-product-locked); delete those if they reappear.
• Upsert standing as knowledge (--kind knowledge, --no-infer). Bare hydradb ingest --text is the memory store (1000-token cap, replace-not-append) — do not put standing there. Use the hydradb CLI with --collection {project}. Do not use Hydra MCP. Never export HYDRADB_COLLECTION globally.
• Keep writes distilled. Skip transient WIP and session checkpoints.
• Do not rewrite personal-collection prefs while hydrating a project.
• Never invent a git sha. Always refresh the waypoint after a successful sync when git is available.
• Preserve Problem / Product / Destination; refresh Current standing from code. Do not strip contracts to fit a single request.

What “in sync” means

Standing reflects current layout, wiring, shipped vs stub, verify commands, still-true locks, dated decisions, scars, and a waypoint at current HEAD.

Workflow

1. Identify project + current git tip

• {project} slug; standing id {project}-standing
• Always pass --collection {project} (do not use Hydra MCP)
• git rev-parse HEAD / --short=12 / --abbrev-ref HEAD
• git log -1 --format=%s and %cI
• git status --porcelain

2. Fetch standing + waypoint

• hydradb inspect {project}-standing --collection {project} only (not a 3–5 hit query)
• Parse ## Sync waypoint → previous commit sha

If missing/empty: recommend /ingest.

3. Decide exploration mode

Validate the previous waypoint sha exists and is an ancestor of HEAD.

4. Explore

Delta-first: git log --oneline <waypoint>..HEAD, git diff --stat and material diffs, dirty paths, dependents/callers, verify scripts if those files changed.

Broader: root manifests, central entrypoints, active product surfaces, CI scripts, added/removed apps/packages/crates.

5. Diff: code vs standing

Focus on material drift: new modules, stubs that became real, contract changes, verify commands, wiring, removed surfaces, overturned locks.

6. Write + advance waypoint

Rebuild the full markdown (same section order as /ingest) and ingest {project}-standing as knowledge with --no-infer:

hydradb ingest --kind knowledge --text "$(cat standing.md)" \
  --source-id {project}-standing --title "{Project} — standing" \
  --no-infer --collection {project}

Always rewrite ## Sync waypoint to current HEAD metadata from step 1.

If leftover sibling ids exist (*-codebase-map, *-decisions-*, *-scars, *-product-locked), delete them.

Waypoint block:

## Sync waypoint

- commit: <full sha>
- short: <12-char sha>
- branch: <branch name>
- subject: <one-line subject>
- committed_at: <ISO from git>
- captured_at: <ISO timestamp of this skill run>

7. Report back

Short: mode, previous → new waypoint, commit range if delta-first, sections changed, corrections (Hydra was wrong → code), skipped items, whether /ingest is recommended.