agentleFS
eunomia-bpfeunomia-bpf/eunomia.dev/.agents/skills/eunomia-community-radar/SKILL.md

eunomia-community-radar

Skill244 starsChanged 5 days agoMITView on GitHub

Monitor approved eBPF, Linux, observability, and AI-infrastructure communities through watchlist-opted-in read-only Slack archives plus ordinary visible browser UI, turn the strongest daily technical question into one source-grounded anonymous eBPF Q&A, and append an anonymized analysis of the day's wider community discussion. Use for daily Slack, Discord, mailing-list, forum, or community monitoring; daily Q&A publication; community-topic summaries; question triage; and maintaining the community watchlist. Do not use for monitoring reactions to already published Eunomia content, GitHub issue or PR patrol, or unsolicited replies.

The file

Show its rules and text

We split this skill into the rules it gives an agent

20 rules. About 0 are common, 13 are distinctive, and 7 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈13

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Keep the watchlist as the allowlist.

    line 25
  2. Distinctive:

    Do not use hidden APIs, internal endpoints, unapproved chat exports, network interception, or scraped message archives for any source.

    line 47
  3. Distinctive:

    Read only approved public channels and threads.

    line 59
  4. Distinctive:

    If a listed channel is inaccessible, record the coverage gap in the run result instead of claiming it was quiet.

    line 61
  5. Distinctive:

    Prefer questions that recur, expose a poorly documented boundary, or connect several discussions.

    line 67
  6. Distinctive:

    When no single message is strong enough, combine related same-day signals into one practical question.

    line 77
  7. Distinctive:

    Verify the answer with official documentation, upstream repositories, standards, kernel documentation, or papers.

    line 95
  8. Very specific:

    After the references, add Community discussion today.

    line 110
  9. Very specific:

    create docs/ebpf-qa/YYYY-MM-DD-<question-slug>.md and its Chinese counterpart.

    line 138
  10. Very specific:

    add the question to docs/ebpf-qa/index*.md.

    line 140
  11. Very specific:

    update .github/publisher/media/community-feedback.md only when a durable technical signal affects future work.

    line 142
  12. Distinctive:

    avoid a separate community report, browsing log, research memo, transcript, or daily artifact.

    line 144
  13. Distinctive:

    Check both routes in a normal browser, including title, references, discussion summary, code, mobile layout, and navigation.

    line 149
  14. Distinctive:

    Do not require a particular local model when the active agent can complete the work.

    line 155
  15. Very specific:

    Before reading anything, verify the connection is transaction_read_only and holds zero admin or write grants.

    line 175
  16. Very specific:

    Use Slack message ts for the 24h and 7d windows.

    line 179
  17. Distinctive:

    No database writes of any kind.

    line 184
  18. Very specific:

    Keep raw text only in a bounded per-run temporary input file with 0600 permissions inside ephemeral isolated OpenCode state, and remove it …

    line 185
  19. Distinctive:

    Never store transcripts, logs, prompts, sessions, or raw text in Git or persistent state.

    line 187
  20. Distinctive:

    Never present a private-community observation as a public quotation or imply that an original participant endorsed the published answer.

    line 202

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Eunomia Community Radar

Turn daily community discussion into one durable answer and one compact view of what eBPF practitioners are trying to solve. Every daily run publishes exactly one combined Q&A and community briefing. Never invent a question or publish without real public primary sources; coverage, freshness, and access gaps are reported honestly and never stop the publication itself.

Required Context

Read before monitoring:

  • CLAUDE.md
  • .agents/README.md
  • .github/publisher/media/community-watchlist.yaml
  • .github/publisher/media/community-feedback.md
  • docs/ebpf-qa/index.md
  • the publication standard

Keep the watchlist as the allowlist. Adding a workspace, server, channel, or forum is a durable scope change and requires an explicit repository edit.

Workflow

1. Review The Full Daily Window

Date every run in America/Los_Angeles and give each daily run its own session and session-scoped state directory. Retries on the same Los Angeles day may continue that day's session and candidate. A published receipt from an earlier date never counts as completing a later day: each later day still requires its own run and its own published Q&A.

Apply the hybrid source policy declared in the watchlist:

  • Slack sources marked archive_opt_in: true use the approved private read-only archive as the primary source and the visible browser as the fallback.
  • Discord and any chat source without an archive remain visible-browser-only.
  • Mailing lists and Reddit use ordinary public browser pages and never APIs.
  • Official documentation may be read in the normal browser.

Do not use hidden APIs, internal endpoints, unapproved chat exports, network interception, or scraped message archives for any source. Archive access must satisfy every invariant in the Archive Access Invariants section.

Review every accessible allowlisted channel for the previous 24 hours. Use visible channel navigation, date dividers, and Page Up, Page Down, Home, or End key presses. The optional visible-channel-review.mjs helper may collect text that is currently rendered in the visible browser and move the same visible message pane with keyboard input. It must never launch a crawler, call a platform endpoint, persist a transcript, or bypass access controls.

Read only approved public channels and threads. Never enter restricted, private, direct-message, customer, or partner channels for editorial mining. If a listed channel is inaccessible, record the coverage gap in the run result instead of claiming it was quiet.

2. Select The Daily Question

Choose the most useful concrete technical question seen in the daily window. Prefer questions that recur, expose a poorly documented boundary, or connect several discussions. The question must be:

  • relevant to eBPF, Linux observability, runtime extension, profiling, security, or adjacent Agent infrastructure;
  • answerable from public primary sources or a reproducible local experiment;
  • useful without the original participant's identity or deployment details; and
  • materially different from an existing Q&A, tutorial, or blog post.

When no single message is strong enough, combine related same-day signals into one practical question. Partial archive coverage, inaccessible or sparse sources, or a thin daily window never stop publication: continue selecting and verifying a real question from existing readable archive material, a genuine unresolved question from the monitored communities regardless of age, permitted public community pages, or public primary documentation, and say in the internal run result which fallback supplied the question. Never publish a placeholder, promotional topic, or speculation, and never fabricate a question or evidence.

3. Write The Answer

Name the article with the anonymized practitioner question. Answer it in detail before discussing the wider community. Write from the technical issue, not the chat wording. Remove names, handles, employers, organization names, timestamps, exact infrastructure, internal URLs, logs, tokens, IP addresses, and distinctive phrasing. Never quote or link a closed-community message on the public page.

Verify the answer with official documentation, upstream repositories, standards, kernel documentation, or papers. Search results can locate sources but cannot replace reading them. Separate established behavior, operational advice, and open limitations. Add a References section after the answer with only the public sources used.

Title/H1 note: keep the H1 free of underscores, backticks, and apostrophes. The publisher renders the question as the page H1 and the validator re-checks that the exact Markdown title string is present verbatim in the live HTML; the site's H1 pipeline strips underscores and HTML-encodes apostrophes, so such characters force a failed live-content check and an extra redeploy. Put helper names as ordinary words in the H1 and reserve backticked code for the answer body.

4. Summarize The Day's Discussion

After the references, add Community discussion today. Cover as much of the daily window as the allowlist and browser access permit. State the number of communities and channels reviewed, then synthesize the main technical themes, where practitioners are getting stuck, and what questions remain unresolved.

Anonymize identity and deployment details without stripping away the technical substance. For every substantive theme, explain the concrete problem or symptom, the likely mechanism or boundary, the practical diagnostic or resolution path, and what remains uncertain. Cite the public primary sources that support these mini-answers. When the daily window contains several real technical discussions, this section should normally be at least twice as detailed as a terse trend-only summary; do not reduce each discussion to one sentence.

Summarize across discussions rather than listing messages. Do not publish participant names, handles, employers, channel names, message links, exact timestamps, private topology, or wording that can be searched back to one person. It is acceptable to say that a channel had no substantive technical discussion. Never treat inaccessible channels as zero activity.

5. Publish The Daily Page

Publish one Q&A per successful run. Before starting a new dated pair, resume any unpublished pair left by an earlier attempt, even after the calendar date changes. A retry or the next day's session must finish that candidate rather than strand it, and must never re-create a page whose article is already published:

  • create docs/ebpf-qa/YYYY-MM-DD-<question-slug>.md and its Chinese counterpart;
  • add the question to docs/ebpf-qa/index*.md;
  • preserve the order: detailed answer, references, community discussion;
  • update .github/publisher/media/community-feedback.md only when a durable technical signal affects future work; and
  • avoid a separate community report, browsing log, research memo, transcript, or daily artifact. The Q&A page is the public daily report.

6. Validate And Publish

Check both routes in a normal browser, including title, references, discussion summary, code, mobile layout, and navigation. Run the repository content tests and build required by CLAUDE.md. Preserve unrelated changes, stage explicit paths, commit on main, merge forward from origin/main when the remote has advanced, and push directly. Unrelated worktree changes are normal in a shared checkout: preserve them, stage and commit explicit owned paths, and continue. Do not require a particular local model when the active agent can complete the work. Keep automation direct: model allowlists, fixed roles, redundant preflights or approvals, hidden orchestration, and gate-style validations that turn coverage gaps, calendar rollover, or unrelated work into publication stoppers are explicitly banned added complexity, as are any new review framework, audit gate, service, scheduler, script, or model change added just to make a run proceed.

Publication is not complete until the change is pushed to main, the deployment finishes, and the public route is verified live. A one-shot print-mode agent that starts a publication must run it to that point before it exits; it must never end immediately after launching an asynchronous validator while expecting a later automatic wake to finish the job.

Archive Access Invariants

For opted-in Slack archive access:

  • Connect only with the DSN from the EUNOMIA_QA_ARCHIVE_DSN environment variable. Never read, print, copy, or persist the DSN in files or state.
  • Before reading anything, verify the connection is transaction_read_only and holds zero admin or write grants.
  • The schema must uniquely match the allowlisted public workspace team name, and every listed channel must exist in it.
  • Use Slack message ts for the 24h and 7d windows; never load or insert time.
  • Query only message text and timestamps for the allowlisted workspace and channels. Do not query users or handles, data JSON columns, or files or attachments.
  • No database writes of any kind.
  • Keep raw text only in a bounded per-run temporary input file with 0600 permissions inside ephemeral isolated OpenCode state, and remove it on exit.
  • Never store transcripts, logs, prompts, sessions, or raw text in Git or persistent state.
  • A missing or ambiguous workspace or channel is inaccessible, not quiet.
  • Public output remains anonymized and source-grounded per the publication standard.

Boundaries

  • eunomia-social-radar owns reactions, citations, comments, and follow-up on content Eunomia has already published.
  • eunomia-community-patrol owns GitHub issues and pull requests across the organization.
  • This skill discovers external technical questions and publishes anonymous Q&A. It does not authorize replies, direct messages, follows, invitations, reactions, or moderation actions unless the user explicitly requests them.
  • Never present a private-community observation as a public quotation or imply that an original participant endorsed the published answer.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.