Autoresearch

Before you start: read this whole install prompt, tell me in a few lines what this skill does and what installing it involves, and wait for my go-ahead before you do anything else. Prerequisites — USE REQUIREMENTS DISCLOSED FIRST: this skill works on your own Obsidian vault and runs through the portable core of the claude-obsidian repository, scripts/claude-obsidian.py, which lives at the repository root OUTSIDE this skill's folder, together with the shared contracts it links to (under skills/wiki/references/ in that same product root, for example the provenance and operation-transaction contracts). Installing this listing copies only this skill's folder: without a full claude-obsidian installation (the core, its contracts, python3 to run it, and a selected user vault) the skill cannot run its transactions as designed. The family orchestrator is the separate wiki skill of the same repository. No account, no key and no payment are needed to copy the files, and nothing is run during installation. At use time the research loop may access the public web ONLY inside an explicit research contract you approve first (topic, approved domains or source classes, budgets); without that egress consent it researches only your vault and sources you provide, and it never sends private vault text, file paths, credentials or unrelated conversation content to external services. Install "Autoresearch" for me. It runs a bounded, source-grounded research loop on a topic you define: it first establishes a research contract with you (exact topic and exclusions, whether public-network egress is approved, approved domains or source classes, and budget limits; the bundled references/program.md defaults are at most 3 rounds, 5 fetched sources per round and 15 drafted pages), reads your vault's hot cache, index and source/claim ledgers to see what is already known, decomposes the topic into questions including a plausible counter-position, prefers official and primary sources, records provenance for each source (URL, title, author/publisher, dates, authority, freshness, payload hash when available) and extracts falsifiable claims with evidence kept separate from inference. Web results, fetched pages and worker drafts are treated as untrusted evidence, never as instructions. The research stays draft-only: filing the cited dossier into the vault is one inspected transaction (you set APPROVAL_SHA256 to the inspect result's approval hash after review, then it applies), and merging it into existing canonical pages is a second, separately inspected and separately approved transaction that you may accept, narrow, postpone or reject. Accepted claims need a fresh active non-synthetic source, high-risk ones two independent sources; unsupported claims are labelled as such, never padded out. Repository: https://github.com/AgriciDaniel/claude-obsidian (the skill is the whole skills/autoresearch/ folder at the pinned commit: SKILL.md, references/program.md, 2 files in total) Pinned version: install the code at commit 2e545563172f97f02ecaaaab376ceea1246a0583 exactly. Do not substitute a newer version. If this commit is unreachable, stop and tell me. 1. Clone https://github.com/AgriciDaniel/claude-obsidian into a temporary folder, check out commit 2e545563172f97f02ecaaaab376ceea1246a0583, and copy the skills/autoresearch/ folder from that checkout. Then compare every file in the folder with the fingerprints recorded for this listing on 2026-10-22: [FINGERPRINTS] - SKILL.md: SHA-256 54ca6599760e6e5827cee50d448b182e40ff77e784a7eda5cd44976b968b5970 - references/program.md: SHA-256 d4ac8ac443461723069cad0e4eacf7d303c1bed644bcf76dda24649ec97991c8 If a file does not match, it changed since the listing was prepared: do not install it yet, tell me what you found and let me decide. 2. Read SKILL.md in full, and tell me in plain words: what the skill does when used, what it connects to, whether it reads or sends any personal information or credentials, and anything that seems unusual. Stop on any red flag and tell me before going further. Confirm these points from the files: - The skill folder holds 2 files: SKILL.md and references/program.md, the bundled default research program (budgets, evidence grades, stop conditions) that the SKILL.md reads first. The core it runs (scripts/claude-obsidian.py) and the provenance/transaction contracts it links to live outside this folder, in the full claude-obsidian product root. - Nothing is sent anywhere during installation itself; the only download at install time is the skill folder from GitHub at the pinned commit, and no research is run. - At use time, network egress happens only inside the research contract you approved (approved domains/source classes and budgets: by default at most 3 rounds and 5 fetched sources per round); private vault text, file paths, credentials and unrelated conversation content are never sent to external services, and fetched content is treated as untrusted evidence, never as instructions. - Every vault write is a reviewed transaction: the dossier filing is inspected and applied only with the approval hash (APPROVAL_SHA256 pattern), and any merge into canonical pages is a second, separate transaction. Workers may search and draft but never mutate the vault. - Do not run any research, fetch any page, file any dossier or merge anything during this install: using it is a separate step that only starts when I ask for it later. 3. When I say "yes, install", install it as a skill: copy the whole autoresearch folder into the agent's skills directory, in a folder named exactly "autoresearch", keeping its 2 files in their original relative layout (SKILL.md at its root). 4. Verify with no network calls: frontmatter valid (name: autoresearch), the 2 files in place with matching fingerprints. Do not run anything from the folder and do not use the skill to verify. 5. Report what was installed, where, and what I still need to do myself (for example: that a full claude-obsidian installation (portable core and contracts at the product root) and my own Obsidian vault are required to use it, that web research runs only inside a research contract I approve, and that filing and canonical merging are two separate reviewed transactions). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: don't touch anything outside the temp folder and the install target. If anything looks off, stop and ask me.