feat: add consume skill (write-side pair to /wiki) #8

Merged
alexion merged 3 commits from feat/consume-skill into main 2026-07-24 20:03:44 -04:00
Owner

Summary

Adds /consume, the write-side counterpart to the already-merged read-only /wiki, completing the pair.
Where /wiki reads the user's Obsidian vault, /consume mines a target for generalized, transferable knowledge and files atomic concept notes into 01 sources/claude/ — the one area of the vault the agent may write, treated as the agent's own wiki.
It never documents a project: a fact earns a note only if it holds on a different target.

Design

  • User-invoked / deliberate (disable-model-invocation: true), run as /consume [target] (defaults to cwd).
  • SKILL.md holds the branch-detection ladder plus the common writing logic; each source type is its own disclosed .md.
  • Reads existing knowledge two ways: /wiki for vault-wide context, plus a direct scan of 01 sources/claude/ for notes to enrich.
  • Plans atomic generalized notes with a many-to-many file→section→note contribution map.
  • Renders an HTML plan to the scratchpad and waits for explicit approval — nothing is written or deleted until the user approves.
  • On approval, writes on the vault's note template, then clears the branch's consumed scaffolding (never before a write).
  • Enforces its own write boundary, since the harness does not apply the vault's permission rules when run from outside it.

v1 branch: project.md

Fires on a directory containing .claude/spec/ and/or .claude/tasks/.
Source code is the truth for the project's current state; .claude/spec and .claude/tasks are the record of why — mined for their reasoning, then deleted.
The ladder is extensible for future source types (YouTube, article, GitHub PR).

Verification

Built via nix build .#consume (auto-discovered, no name collision).
A dry-run subagent drove the skill to the approval gate: branch detection, the generalizes test, and the contribution map all worked; findings on cleanup-scope wording, spec framing, granularity, and no-note files were folded back in.
The live write + delete half is first exercised on a real run, which pauses at the same approval gate before any mutation.

## Summary Adds `/consume`, the write-side counterpart to the already-merged read-only `/wiki`, completing the pair. Where `/wiki` reads the user's Obsidian vault, `/consume` mines a target for **generalized, transferable knowledge** and files atomic concept notes into `01 sources/claude/` — the one area of the vault the agent may write, treated as the agent's own wiki. It never documents a project: a fact earns a note only if it holds on a *different* target. ## Design - **User-invoked / deliberate** (`disable-model-invocation: true`), run as `/consume [target]` (defaults to cwd). - `SKILL.md` holds the branch-detection ladder plus the common writing logic; each source type is its own disclosed `.md`. - Reads existing knowledge two ways: `/wiki` for vault-wide context, plus a direct scan of `01 sources/claude/` for notes to enrich. - Plans atomic generalized notes with a many-to-many file→section→note contribution map. - Renders an **HTML plan to the scratchpad and waits for explicit approval** — nothing is written or deleted until the user approves. - On approval, writes on the vault's note template, then clears the branch's consumed scaffolding (never before a write). - Enforces its own write boundary, since the harness does not apply the vault's permission rules when run from outside it. ## v1 branch: `project.md` Fires on a directory containing `.claude/spec/` and/or `.claude/tasks/`. Source code is the truth for the project's current state; `.claude/spec` and `.claude/tasks` are the record of *why* — mined for their reasoning, then deleted. The ladder is extensible for future source types (YouTube, article, GitHub PR). ## Verification Built via `nix build .#consume` (auto-discovered, no name collision). A dry-run subagent drove the skill to the approval gate: branch detection, the *generalizes* test, and the contribution map all worked; findings on cleanup-scope wording, spec framing, granularity, and no-note files were folded back in. The live write + delete half is first exercised on a real run, which pauses at the same approval gate before any mutation.
alexion added 2 commits 2026-07-24 20:03:00 -04:00
A model-invoked skill that answers questions from the user's Obsidian
vault, read-only, over the filesystem. Enforces the vault's own
settings.json deny-Read globs itself, since the harness does not apply
them when running from another directory. Uses the 02 tags/ hubs as a
recall backstop and reads to saturation, then answers with per-note
provenance.
/consume is the write-side counterpart to the read-only /wiki: it mines a
target for generalized, transferable knowledge and files atomic concept
notes into 01 sources/claude/, the agent-owned area of the Obsidian vault.

It is user-invoked/deliberate. SKILL.md holds the branch-detection ladder
plus the common writing logic — read existing knowledge (via /wiki and a
direct scan), plan atomic generalized notes with a file-to-section-to-note
contribution map, present an HTML plan and wait for explicit approval, then
write on the vault's note template and clear the consumed scaffolding.

v1 ships one branch, project.md: source code is truth for current state,
while .claude/spec and .claude/tasks are the record of why (mined, then
deleted). The ladder is extensible for future source types.
alexion added 1 commit 2026-07-24 20:03:35 -04:00
alexion merged commit 8c17f3aeb8 into main 2026-07-24 20:03:44 -04:00
alexion deleted branch feat/consume-skill 2026-07-24 20:03:44 -04:00
Sign in to join this conversation.
No Reviewers
No Label
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: alexion/skills#8