Skip to content

feat: factor spec/plan/design metadata files into the wiki, not just code #17

Description

@AvyanshKatiyar

Problem

almanac init and capture build pages from code and subsystems, but they largely ignore the planning/spec/intent files that sit alongside the code — design docs, DEPLOY_PLAN.md, *-execution-plan.md, specs/, notes/ folders, roadmap PDFs. These describe intended and future work — which is precisely "what the code can't say," the project's own tagline. A wiki-trusting agent therefore can't find them.

Concretely, on a fresh 15-page init of a multi-project directory:

  • almanac search "deploy plan execution roadmap"0 results, even though a DEPLOY_PLAN.md and an execution-plan markdown both exist in the tree.
  • Page sources: blocks list code files (package.json, server/index.ts, route files…) but no spec/plan/notes files, so their content is never indexed for full-text search.
  • --mentions only resolves these files by folder (via the parent project's page); loose plan docs, notes/ dirs, and roadmap PDFs that aren't under a "notable" subsystem get no page and no content capture at all.

So intent documents — the part a new agent most needs to not duplicate or contradict planned work — are the least discoverable thing in the wiki.

Proposed shape

When building/capturing, treat spec & planning metadata as first-class source material:

  • Recognize files/dirs by name signal: spec, plan, design, roadmap, notes, TODO, RFC, CHANGELOG, docs/, and conventional spec paths (e.g. docs/**/specs/*.md).
  • At minimum, index their content (so full-text search hits them) and add them to the relevant page's sources:.
  • Where a plan is substantial and standalone, give it its own page (e.g. a roadmap/planned-work page) rather than folding it silently into a code page.

Smallest useful version: don't skip .md plan/spec files during the source-selection pass, and capture their text into the index.

Alternatives considered

  • Page-per-file / "index everything" — rejected; it fights the notability bar and drowns signal in noise.
  • Manually authoring plan pages — works, but doesn't scale and is wiped/overwritten on re-init; the gap is structural, not per-repo.

Who benefits?

Any agent picking up planned-but-unbuilt work, and any solo dev/researcher whose repo or working directory holds roadmaps, design specs, and notes the code doesn't reflect yet. Today those people get a wiki that describes only what's already built.

Extra context

Tagline is "captures what the code can't say: decisions, flows, invariants, gotchas." Plans and specs are the strongest example of that, yet they're currently the most likely to be skipped.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions