Skip to content
Agentmart
Submit a tool

curate-apply

Review the background curator's latest nightly proposals and apply the ones the user approves.

Skill by julianhernandez2155

IndexedMIT
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/julianhernandez2155-curate-apply

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Review the background curator's latest nightly proposals and apply the ones the user approves. The curator (the `curator` skill, run on a schedule) is PROPOSE-ONLY and writes a report; THIS skill is the human-gated apply step. Trigger on `/curate-apply`, "review the curator", "apply curator proposals", "what did the curator find", "go through last night's proposals", or when the user wants to act on the nightly report. INTERACTIVE only — never auto-applies; always presents the list and applies only the explicitly selected items.

Use cases: documents

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/curate-apply

What's inside

Skill name
curate-apply
Description
Review the background curator's latest nightly proposals and apply the ones the user approves. The curator (the `curator` skill, run on a schedule) is PROPOSE-ONLY and writes a report; THIS skill is the human-gated apply step. Trigger on `/curate-apply`, "review the curator", "apply curator proposals", "what did the curator find", "go through last night's proposals", or when the user wants to act on the nightly report. INTERACTIVE only — never auto-applies; always presents the list and applies only the explicitly selected items.
License
MIT
Instructions
53 lines in SKILL.md · view on GitHub

Files (1)

  • SKILL.md

SKILL.md (first lines)

# curate-apply — review and apply the curator's proposals

The `curator` skill runs unattended and is hard-blocked from writing anything but its own report. This skill is the other half of the loop: you (the user, present and interactive) review those proposals and apply the ones worth applying. The apply step is deliberately human-gated — that's the whole reason the curator never auto-applies.

## Step 1 — locate the latest report

Newest report dir under `~\curator\proposals\` (sort by date). Read `nightly-report.md` there.
- If none exists: say so, suggest running the curator (`powershell -File ~/.claude\scripts\curator-run.ps1 -Mode nightly`), and stop.
- If the report already ends with an `## Applied — <date>` section: tell the user it's been partly processed and show what's left, so nothing is double-applied.

## Step 2 — present the proposals as a selectable checklist

Parse the report's sections (synthesis, context hygiene, KB lint, sync check). Render a compact numbered list — one line per actionable proposal — grouped by section, each showing: number, what it does, and the target file. Example shape:

```
SYNTHESIS
 1. Create wiki/prompts/anthropic-prompting-patterns.md from raw/2026-05-28-prompting-best-practices.md
 2. Write wiki/patterns/stable-purpose-layer.md (referenced in memory, missing)
HYGIENE
 3. Add FROZEN banner to example-pipeline/docs/STATUS.md
 4. Create su-kb-site/docs/log/2026-05-28.md + 2026-06-01.md
 5. Move "example-eval skill" from TODO Active to Completed
LINT
 6. 3 wiki notes missing provenance: — list
```

Then ask the user which to apply. Accept flexible selection: explicit numbers ("1, 3, 5"), ranges, by group ("all the hygiene ones"), "all", or "none". **Do not proceed without an explicit choice.** Pure-information items (lint lists, sync flags with no concrete fix) are shown but marked "review only — no auto-apply."

## Step 3 — apply only the selected items

For each approved item, do the real change in the live Workspace (normal interactive edits — these are NOT propose-only; you have the user's explicit go):
- **Synthesis notes:** before creating a *new* file, confirm the report's absorb-before-create verdict — prefer **Improve/Absorb** (append to an existing note) over **Save** (new file); only create new if the report justified why it isn't an append. When creating: use the report's proposed frontmatter. Honor the Team KB convention — `provenance:` (+`source:`) is mandatory; add the suggested `[[wikilinks]]`; if the body needs real content, read the raw source first and write a genuine synthesis, don't stub it.
- **STATUS / log / ADR changes:** prefer routing through the `decision-log` skill conventions (don't hand-roll a format decision-log owns). Apply the report's proposed diff faithfully; show the edit.
- **TODO/index/INSIGHTS reconciliation:** apply the specific line change shown.
- **Memory changes:** apply consolidations/updates to `~/.claude/projects/.../memory/` per the proposal; keep the `MEMORY.md` index in sync.

Show each change before/after (or the diff). If an item turns out to be wrong on closer inspection (the curator read stale state, or you disagree), say so and skip it — the curator proposes, your judgment decides.

## Step 4 — mark what was applied

Append a short `## Applied — <date>` section to the report listing which item numbers were applied and which were skipped (and why, briefly). This makes the report idempotent — a re-run of this skill won't re-offer applied items. Writing to the report file (under `~\curator\`) is fine; it's not protected.

**Rejected-ledger (companion to the curator's diffing).** For each proposal the user explicitly **declined** (not merely "review-only" info items), append one line `{id} | {reason} | {date}` to `~\curator\rejected-ledger.md` (create it if absent; use `Get-Date -Format 'yyyy-MM-dd'`). The nightly curator reads this ledger and **suppresses** already-rejected proposals, so a declined item doesn't keep resurfacing every night. Only log genuine rejections — not items deferred for later.

## Discipline

- **Never auto-apply.** Selection is always explicit.
- **Trust your eye over the report.** It was generated from a point-in-time read and may be stale (the curator itself flags state as "verify and apply").
- **Respect the conventions, don't reinvent them.** Provenance on every wiki note; decision-log owns STATUS/log/ADR shape; memory index stays consistent.
- **Literal-flag activation.** A flag/option is active ONLY if its literal token appears in the invocation args; a flag merely *documented* in a report or note is not active. Don't treat "the report mentions `--foo`" as "`--foo` is on" when applying a proposal.
- **Batch related edits, show diffs, keep the user in the loop** — this is a quick morning triage, not a long ceremony.

From julianhernandez2155/claude-code-harness, MIT.

More from julianhernandez2155/claude-code-harness

Signals

Stars
0
Forks
0
Open issues
0
Last commit (this folder)
Jun 8, 2026 (4 months ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from julianhernandez2155

Cite this page

curate-apply skill. Agent Mart. https://agentmart.dev/skills/julianhernandez2155-curate-apply (last updated 2026-10-24)