openbrief

name	OpenBrief
description	Use OpenBrief for local-first brief runs through the installed OpenBrief JSON runner. Reject direct SQLite reads or writes as runner JSON substitutes, HTTP/MCP/source-built runner bypasses, and unreviewed imports of delivery history, latest-seen state, run state, or inferred private configuration. User-pointed legacy automation/config migration may inspect only the named input, draft OpenBrief sources and outlet policies for review, and apply after approval through openbrief config.
license	MIT
compatibility	Requires local filesystem access and an installed OpenBrief binary on PATH.

OpenBrief

Use this skill for routine local OpenBrief brief runs and configuration tasks. The production interface is AgentOps: this SKILL.md plus the installed JSON runner.

openbrief config
openbrief brief

Pipe exactly one JSON request to one runner command, then answer only from the JSON result. The configured local database path is already available through the environment. For routine requests, do not pass --db unless the user explicitly names a specific dataset.

The runner honors OPENBRIEF_DATABASE_PATH. The database stores brief sources, outlet policy rows, latest-seen state, health warnings, delivery records, and recent sent items. Do not maintain repo-local state files.

Reject Before Tools

Answer with exactly one assistant response and no tools when the user asks to perform a production OpenBrief task by bypassing the installed runner. Reject requests to:

• read from or write to SQLite directly as a substitute for runner JSON
• use HTTP, MCP, or source-built command paths instead of the installed runner
• import delivery history, latest-seen state, run state, or inferred private configuration without user review

For unsupported production workflows, say the production OpenBrief runner does not support that workflow yet.

Allowed Contexts

Repository development, docs updates, tests, release verification, security review, and migration design may inspect repository files.

Legacy Migration

Legacy automation/config migration is allowed when the user explicitly points to the input. Inspect only the named input, draft OpenBrief sources and outlet policies for review, and apply only after approval through openbrief config.

Do not import delivery history, latest-seen state, run state, or inferred private configuration without user review.

Do not run openbrief --help, command -v openbrief, repo searches, broad file enumeration, or source inspection for routine tasks. Use the request shapes below.

Config Tasks

Run configuration tasks with:

openbrief config

Common request shapes:

{"action":"init"}
{"action":"inspect_config"}
{"action":"replace_sources","sources":[{"key":"example","label":"Example","kind":"rss","url":"https://example.com/feed.xml","section":"technology","threshold":"medium","enabled":true,"url_canonicalization":"none","outlet_extraction":"none","dedup_group":"news","priority_rank":10,"always_report":false}]}
{"action":"upsert_source","source":{"key":"example","label":"Example","kind":"rss","url":"https://example.com/feed.xml","section":"technology","threshold":"medium","enabled":true}}
{"action":"upsert_source","source":{"key":"tool-releases","label":"Tool Releases","kind":"github_release","repo":"owner/name","section":"releases","threshold":"always","enabled":true}}
{"action":"delete_source","key":"example"}
{"action":"replace_outlet_policies","outlets":[]}
{"action":"set_brief_options","max_delivery_items":7}

Supported source kinds are rss, atom, and github_release. Supported thresholds are always, medium, high, and audit. A fresh database has no sources.

Optional generic feed-processing fields:

• url_canonicalization: none, feedburner_redirect, or google_news_article_url
• outlet_extraction: none, title_suffix, url_host, or rss_source
• dedup_group: string used to collapse same-topic candidates across sources
• priority_rank: lower numbers rank earlier when duplicate candidates compete
• always_report: true makes new feed items must_include

Feed URLs, topics, priorities, always-report settings, outlet policies, and brief options are operator configuration. max_delivery_items controls the normal total delivery count and defaults to 7 when unset. User-directed legacy migration can draft sources and outlet policies for review, but operational state must not be imported without a dedicated runner-backed workflow.

Brief Tasks

Run brief tasks with:

openbrief brief

Request shapes:

{"action":"validate"}
{"action":"run_brief","dry_run":false}
{"action":"record_delivery","run_id":"run_id_from_run_brief","message":"- [Title](<https://example.com>)"}

For run_brief, use must_include, candidates, max_delivery_items, health_footnote, previous_briefs, suppressed_recent, suppressed_policy, suppressed_unresolved, suppressed, and fetch_status from the JSON result. Do not inspect any files to supplement the result. If the result has rejected: true, answer with the rejection_reason.

Current brief body rules:

• Include all must_include items first.
• Fill remaining slots up to max_delivery_items total bullets from candidates using brief judgment appropriate to each candidate's section and threshold.
• If must_include exceeds max_delivery_items, include all must_include items and no candidates.
• Use suppressed hyperlinks only: - [Title](<https://example.com>).
• If health_footnote is non-empty, append it as plain text after bullets.
• If there are no bullets and no health footnote, the current brief body is exactly NO_REPLY.
• Before the final answer, call record_delivery with the exact current brief body, including NO_REPLY. Do not include prior brief history in the recorded message.

Final answer rules:

• If the record_delivery result includes final_answer, answer with exactly that string and ignore run_brief.previous_briefs.
• record_delivery.final_answer contains the latest three delivery records from the runner-owned delivery table: Current brief for the just-recorded delivery, then up to two Previous brief (<delivered_at>) sections.
• If record_delivery.final_answer is absent, fall back to the legacy run_brief.previous_briefs rules: if previous_briefs is empty, answer with only the current brief body; otherwise answer with a Current brief heading, the current brief body, then up to two previous sections from previous_briefs in JSON order.
• In each previous section, render that entry's message exactly as recorded. Do not summarize, paraphrase, strip links, or turn previous entries into prose such as "Delivered 7 items, including ...". Preserve Markdown links, NO_REPLY, and any health footnote text in the recorded previous message.

Validation rejections are JSON results with rejected: true. Runtime failures exit non-zero and write errors to stderr.