Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Call the Obsidian Local REST API directly (over HTTP) for vault operations the mcp__obsidian__* tools do NOT expose — move/rename a note, overwrite a whole file atomically (PUT), act on the currently-open active file, run an Obsidian command, open a note in the UI, list all tags, or do date-specific periodic-note CRUD. Prefer the mcp__obsidian__* tools for plain read/append/patch/delete/search; fall back to this skill only when the required method is missing from MCP.
.claude/skills/davepoon-obsidian-rest-api/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-13 | ✗→✓ | ▲ Improved | -18% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 5% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 269% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 90% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -25% | 0% |
The connected obsidian MCP server exposes only a subset of the Obsidian Local REST API (plugin obsidian-local-rest-api). This skill provides the full API surface plus an authenticated request wrapper, so a missing MCP method is called over HTTP instead of being worked around with hacks (e.g. delete+recreate to rename a note).
Use the mcp__obsidian__* tools first for read, append, patch, delete, and search. Fall back to this skill only for operations that have no MCP tool:
config (~/.claude.json) or OBSIDIAN_* env vars — no hardcoded secrets.
references/api_reference.md),with the header enums (Operation, Target-Type, Target-Scope), the custom MOVE contract, and the search (JsonLogic/Dataview) formats.
Call the wrapper scripts/olrapi.sh <METHOD> <path> [curl args...]:
bashS=scripts/olrapi.sh # adjust to the skill's install path # rename/move a note (the most common reason to reach for this skill) "$S" MOVE "/vault/Path/To/Old Name.md" -H 'Destination: Path/To/New Name.md' # move into a folder, keeping the filename (trailing slash on Destination) "$S" MOVE "/vault/Inbox/todo.md" -H 'Destination: Archive/' # atomically overwrite a whole note "$S" PUT "/vault/Path/Note.md" -H 'Content-Type: text/markdown' --data-binary @/tmp/body.md # read a note as structured JSON (frontmatter + tags + stat) "$S" GET "/vault/Path/Note.md" -H 'Accept: application/vnd.olrapi.note+json' # list tags, run a command, open a note in the UI "$S" GET /tags/ "$S" POST "/commands/editor:toggle-bold/" "$S" POST "/open/Path/Note.md?newLeaf=true"
The wrapper prints <<HTTP nnn>> after the body. Success: 200/204. On MOVE, 409 means the destination exists — add -H 'Allow-Overwrite: true' to force. For non-trivial calls, load references/api_reference.md.
{filename} is vault-relative (no leading slash on the vault path).MOVE Destination header(e.g. r%C3%A9sum%C3%A9.md). Destination rejects absolute (/…) paths.
Target-Type (heading|block|frontmatter)+ Target headers on GET/PATCH/POST.
User: "Rename 3-Resources/Draft.md to 3-Resources/Final.md in my vault."
Output:
bashscripts/olrapi.sh MOVE "/vault/3-Resources/Draft.md" \ -H 'Destination: 3-Resources/Final.md' # <<HTTP 204>> — moved, history preserved, internal links updated
scripts/olrapi.sh GET /openapi.yaml. GET / shows the plugin version.
PUT over delete+recreate for whole-file overwrites — it is atomic andkeeps the file's identity.
curl -k) and HTTP on 27123.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-13 | fail→pass | 9,761 | 3,274 | -66% | 1 | 1 | 0% | 1,822 | 1,500 | -18% | 0 | 0 | — |
case-01 | fail→fail | 19,303 | 7,932 | -59% | 1 | 1 | 0% | 3,446 | 1,681 | -51% | 0 | 0 | — |
case-02 | fail→fail | 8,162 | 4,259 | -48% | 1 | 1 | 0% | 1,819 | 1,210 | -33% | 0 | 0 | — |
case-03 | fail→fail | 5,198 | 7,402 | +42% | 1 | 1 | 0% | 905 | 1,230 | +36% | 0 | 0 | — |
case-04 | fail→pass | 8,562 | 2,597 | -70% | 1 | 1 | 0% | 1,397 | 1,469 | +5% | 0 | 0 | — |
case-05 | fail→pass | 3,566 | 3,087 | -13% | 1 | 1 | 0% | 417 | 1,537 | +269% | 0 | 0 | — |
case-06 | fail→pass | 5,195 | 3,425 | -34% | 1 | 1 | 0% | 835 | 1,587 | +90% | 0 | 0 | — |
case-07 | fail→pass | 10,545 | 2,653 | -75% | 1 | 1 | 0% | 1,970 | 1,484 | -25% | 0 | 0 | — |
case-08 | fail→pass | 7,285 | 3,504 | -52% | 1 | 1 | 0% | 1,206 | 1,345 | +12% | 0 | 0 | — |
case-09 | pass→pass | 11,664 | 4,507 | -61% | 1 | 1 | 0% | 2,159 | 1,879 | -13% | 0 | 0 | — |
case-10 | pass→pass | 10,669 | 6,091 | -43% | 1 | 1 | 0% | 1,978 | 2,028 | +3% | 0 | 0 | — |
case-11 | fail→pass | 8,935 | 3,068 | -66% | 1 | 1 | 0% | 1,388 | 1,498 | +8% | 0 | 0 | — |
case-12 | fail→pass | 8,480 | 2,442 | -71% | 1 | 1 | 0% | 1,554 | 1,287 | -17% | 0 | 0 | — |
case-14 | pass→pass | 5,723 | 4,031 | -30% | 1 | 1 | 0% | 998 | 1,739 | +74% | 0 | 0 | — |
case-15 | fail→pass | 8,042 | 2,538 | -68% | 1 | 1 | 0% | 1,347 | 1,481 | +10% | 0 | 0 | — |
case-16 | pass→pass | 7,275 | 2,331 | -68% | 1 | 1 | 0% | 1,221 | 1,403 | +15% | 0 | 0 | — |
case-17 | pass→pass | 10,015 | 1,991 | -80% | 1 | 1 | 0% | 1,967 | 1,217 | -38% | 0 | 0 | — |
case-18 | fail→pass | 8,327 | 2,633 | -68% | 1 | 1 | 0% | 1,357 | 1,404 | +3% | 0 | 0 | — |
case-19 | fail→pass | 11,320 | 1,611 | -86% | 1 | 1 | 0% | 1,767 | 1,257 | -29% | 0 | 0 | — |
case-20 | pass→pass | 3,880 | 3,184 | -18% | 1 | 1 | 0% | 634 | 1,513 | +139% | 0 | 0 | — |
case-21 | pass→pass | 6,688 | 4,962 | -26% | 1 | 1 | 0% | 1,169 | 1,643 | +41% | 0 | 0 | — |
case-22 | fail→pass | 6,128 | 3,832 | -37% | 1 | 1 | 0% | 947 | 1,580 | +67% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 19 counted toward the lift figure. The other 3 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +55 percentage points is the difference between those two pass rates over the 19 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.