All features, in detail

This page holds the detailed feature guides that used to make the README long. Nothing here is needed to get started: begin with the Start here guide and the quickstart. Come back when you want a specific capability.

Every feature below is offline and draft-only unless it says otherwise. None of them publishes anything, and none of them promises views, followers or growth. Words you do not know are explained in the glossary.

Jump to: Growth actions and analytics · International channels · Profile and post audits · Creative preflight and creator amplification · Supported operations · Campaigns and experiments · Media checks · Scheduling · Choosing networks · DEV Community

Evidence-ranked social growth actions (offline, Issue #6)

reachweaver suggest combines individual profile fixes, native post-metric coverage, unresolved publishing-safety incidents, owner-exported engagement triage, SEO destination conflicts and preregistered experiment history into one evidence-ranked, human-review-only task list. Safety and privacy issues come first. It never guesses account rankings or predicts likes/impressions.

reachweaver suggest --snapshot examples/profile-post-audit.example.json --json
# Optional: add owner-exported data from an EXISTING private DB; no DB created or modified
reachweaver --db state/reachweaver.sqlite suggest --include-db --snapshot state/owned-profiles.json --limit 15

reachweaver weekly-digest --out state/digest.md takes the same inputs and writes them as one private Markdown file; nothing is sent. See the next-actions runbook. Real native account analytics need an owner-approved export or a separately authorized collector; suggest itself never connects to a platform, changes the queue, drafts, replies, likes, schedules or publishes.

Consent-aware native social analytics from local CSV (offline)

Profiles and content audits can now incorporate your own source-labeled native metrics, without connecting social accounts or modifying databases. Export the metrics you are authorized to use, normalize the columns once, and run:

reachweaver post-audit --file examples/profile-post-audit.example.json \
  --metrics-csv examples/profile-post-metrics.example.csv --owner-authorized --json
reachweaver suggest --snapshot examples/profile-post-audit.example.json \
  --metrics-csv examples/profile-post-metrics.example.csv --owner-authorized --json

The CSV is read into memory, never published or saved elsewhere; owner acknowledgement is not independent API verification. Blank metrics remain unknown and duplicate posts are rejected. See native CSV import contract and individual platform audits.

International social discovery (research only, issue #64)

The offline distribution planner catalogs 110 channels across 18 market pilots, including India-focused ShareChat, Moj, Josh, Chingari, Hipi, Public App, Kutumb, Arattai Channels, LocalCircles, Pratilipi and Rooter; plus Xiaohongshu/REDnote, Weixin Channels, Naver, VK and other regional platforms. See regional channel research and India use cases and community, creator and publishing platform research and try reachweaver distribution-plan --region india --goal awareness --limit 12 --json. New entries are manual-only with unverified native scheduling; community discussion and opt-in channel messages stay human-only. No new API adapter, account connection or remote publication is implied.

Per-platform profile and post audits (offline, Issue #3)

ReachWeaver now audits each network on its own terms. LinkedIn headlines and featured proof, YouTube channel organization and retention, Instagram original Reels and reach, DEV technical articles, Mastodon profile metadata and verified links, Reddit rules, Product Hunt maker stories, and Chinese developer channels receive different profile checklists and native content experiments. Twenty-five platform and community surfaces are described in the complete audit playbook. This is not a universal ranking formula or a guarantee of likes, impressions, views or follows.

# Pure offline example: no account, API keys, website fetch, local DB or changes.
reachweaver profile-audit --file examples/profile-post-audit.example.json --json
reachweaver post-audit --file examples/profile-post-audit.example.json --age-hours 168 --min-posts 3 --json

# For a REAL audit, replace the example with your OWN owner-confirmed profile
# observations and source-labeled native analytics kept locally under state/.
reachweaver profile-audit --file state/owned-profiles.json --platform linkedin --json
reachweaver post-audit --file state/owned-profiles.json --platform youtube --json

# Peer benchmark from public counters you copied by hand (no fetching, nothing stored; docs/PEER-BENCHMARK.md)
reachweaver peer-benchmark --file examples/peer-benchmark.example.json

The profile score is only an observed checklist-completion heuristic and is omitted when less than 60% of the platform’s fields have been checked. Post rates require compatible format, account, platform, native metric source and observation age; missing data is unknown. Inputs include no passwords or cookies and the audit commands never log in, change a profile, draft, schedule, publish, follow or generate votes. Without an actual owner-supplied snapshot, the skill offers a general playbook, not an audit of your real accounts. See the portable skill reference.

Creative preflight and creator amplification (offline)

creative-preflight reads one local brief with 1-6 draft posts and prints a disclosed editorial checklist per draft: audience and topic terms, opening length, owner-supplied source links, the platform’s listed formats, captions or alt text, and the call to action, plus the hashtag, length and duplicate findings of seo-content-check. Unknown inputs lower assessed_weight instead of counting as failures. It is not a ranking formula or a reach prediction, and human-only destinations (Reddit, Hacker News) get no generated prompts.

creator-match, creator-kit and creator-coverage help you prepare an original post so that relevant creators can choose to cover, critique, reproduce or cite it. You supply every creator entry yourself; nothing is looked up or scraped. Opted-out or declined creators are always held. Each row says that ReachWeaver contacts nobody, Reddit rows add the human-only rule, and the kit carries consent and disclosure reminders. Coverage counts only owner-confirmed, independent pickups.

reachweaver creative-preflight --file examples/creative-preflight.example.json
reachweaver creator-match --file examples/creator-candidates.example.json
reachweaver creator-kit --file examples/creator-source.example.json --out ./creator-kit-review
reachweaver creator-coverage --file examples/creator-coverage.example.json --json

See creative preflight and creator amplification.

Developer attention and contributor growth

For a source-backed, moderation-aware 30-day plan prioritizing DEV, LinkedIn, GitHub-native discovery and Bluesky over indiscriminate cross-posting, see the open-source developer growth playbook. It covers first-time-contributor onboarding, platform rules, Reddit/HN recovery, qualified-traffic measurement and the changed Hacktoberfest 2026 rules. It is strategic research, not evidence of published posts or promised reach.

Claude Code growth workflow: The installed skill now loads its contributor-growth reference when you ask about GitHub visitors/contributors or restricted HN/Reddit promotion. Example: /reachweaver — Grow first-time contributors for YazSes using DEV, LinkedIn, GitHub and Bluesky over the next 30 days. Keep Hacker News human-only and Reddit community-compliant. Show the measurable checklist; do not publish anything without approval.

Worldwide audience selection: executable, offline research

ReachWeaver now includes a versioned and tested research registry spanning 100 networks/surfaces and 18 country/global market profiles, including India, China, Japan, Southeast Asia, Africa and Brazil. It also records native scheduling evidence separately from ReachWeaver’s actual account and API capabilities. Results are editorial channel hypotheses, never guaranteed traffic, a detected account connection, a remote scheduling receipt or publishing permission.

# No account credentials or network access needed. These NEVER queue or publish.
python3 -m reachweaver.cli distribution-plan --region india --goal adoption --limit 5
python3 -m reachweaver.cli distribution-plan --region china --goal contributors --json
python3 -m reachweaver.cli distribution-plan --region nigeria --goal collaboration --date 2026-10-14 --json
# A worldwide audience has no one local timezone:
python3 -m reachweaver.cli distribution-plan --region global --goal awareness --date 2026-10-14 --timezone Europe/Rome --json

Use --goal contributors|adoption|awareness|collaboration. Candidate local lunchtime/evening windows are illustrative tests, not verified optimal hours; the output includes offset-aware local and UTC timestamps. Hacker News is a special human-authored, manual-only channel and must not receive AI-written or AI-edited submission content.

Read global strategy, operating/governance standard, platform algorithm and scheduling evidence, developer contributor playbook, and the owner-reported activity inventory. The canonical machine-readable data is reachweaver/distribution_registry.json; the complete standalone Claude Code skill bundles that registry and the planner.

Supported operations

Capability	v0.1
Strategic research and campaign creation via Claude Code	Router, specialized creative and evidence references + templates
Worldwide region/goal channel planner	Offline tested registry, explicit native/API/ReachWeaver status, optional IANA-zone experimental slots; NO posting
Distinct content drafts across 12 platforms	Skill instructions + conservative text lint
Account inventory / missing-credential check	Local config only; no interactive OAuth UI
Personas (work, family), a username per network, wrong-account refusal	profile, account-route and --persona on draft/check/approve; offline, mock-tested (guide)
Sign-up details for accounts you create yourself, in the platform’s language	account-kit: one plan, one approval, local packets; creates no account, fills no form, sees no password or code (guide)
Authenticated account identity preflight	Explicit online check for Bluesky, Mastodon, Telegram and LinkedIn (/v2/userinfo sub); mock-tested, no account linked. Checks for Instagram, TikTok, YouTube, Threads and Facebook exist but are refused before any credential read while those adapters are disabled
One-post supervised pilot isolation	show POST_ID, publish --post-id POST_ID, remote identity check before send
Approval, scheduling, dry-run	Functional local CLI
Safe persistent outbox, kill switch, reconciliation	SQLite-backed local CLI
Bluesky / Mastodon / Telegram	Text-only live API adapters, mocked tests; one real text post each verified for Bluesky and Mastodon (receipts), none for Telegram. Mastodon image attachments (#13) and Bluesky/Mastodon reply sending (#17) are mock-tested but registry-disabled until a live receipt. Bluesky: 300-grapheme check, link/hashtag facets, optional custom PDS service, deterministic record key, verified uri/cid receipt. Mastodon: pinned per-instance max_characters, visibility (no DMs), optional language/spoiler_text, verified receipt. See PLATFORMS.md
LinkedIn personal profile (member)	Text-only Posts API adapter with pinned Linkedin-Version, mocked tests; organization pages unsupported (guide)
Threads, Facebook Page, Instagram, YouTube, TikTok	Mock-tested adapters (Threads text #19, Facebook Page text/link #21, Instagram single image #20, YouTube upload #22, TikTok creator-inbox video #23), owner-approved 2026-10-09 and disabled in the capability registry until a supervised live receipt; manual-export creator packages meanwhile
X, Reddit, Pinterest	Creative/manual workflow only; no adapter
Weibo, Xiaohongshu, Douyin, Bilibili, WeChat Official Accounts, Zhihu and six more Chinese platforms	China-aware manual-export packages: sourced limits, platform topic syntax, AI-label and ad-label decisions; you post by hand (CHINA.md). No adapter: no open posting API the owner qualifies for
Analytics / comparisons	User-supplied snapshot ingestion, read-only polling of public Bluesky/Mastodon counters (metrics-fetch), coverage-aware local report and matched-age directional experiment evaluator
Campaign validation and brand identity boundaries	Offline evidence/brand JSON checks; immutable idempotent draft staging
Private/owner analytics fetching (views, impressions, clicks) or media rendering	Not implemented; metrics-fetch reads only public like/repost/reply(/bookmark) counters
Cloud scheduler with cross-device durable state	Not implemented; see persistent cron guidance

New in v0.2: source-bound campaigns and evidence-aware learning

Instead of letting a model produce untraceable copy and automatically broadcasting it, build a brand-specific, evidence-registered campaign package. The offline validator checks required source declarations, private brand-to-account boundaries, unsupported media formats and near-identical cross-channel text. It cannot independently establish that a webpage supports the claim; Claude/human editorial verification is still required.

# Works even before accounts or API credentials are connected
python3 -m reachweaver.cli campaign-check --file examples/campaign.reachweaver.json --brand examples/brand.reachweaver.json --json
python3 -m reachweaver.cli campaign-export --file examples/campaign.reachweaver.json --brand examples/brand.reachweaver.json --out /tmp/reachweaver-manual-review

# Requires your ignored local config/accounts.json and a reviewed brand account allowlist.
# Repeated calls with an identical manifest return the same draft IDs; changing a staged manifest is rejected.
python3 -m reachweaver.cli campaign-stage --file examples/campaign.reachweaver.json --brand examples/brand.reachweaver.json

# After human approval + separately authorized real publishing, import native metric snapshots:
python3 -m reachweaver.cli experiment-evaluate --file examples/experiment.example.json

Preregister before publishing: experiment-preregister examples/growth-plan.example.json --out FILE rejects more than 3 arms, multi-dimension or near-duplicate variants, vanity primary metrics (likes, views) and early stopping on a win, and it requires notes acknowledging six confounders. experiment-verify PLAN --hash H then shows whether the plan was edited after registration. experiment-readout --prereg FILE --posts MAP.json verifies the hash, maps arms to published posts and reads out against the local database without writing to it. Readouts are labeled only insufficient_data, inconclusive or directional, never proven. Each readout states whether the design was owner-declared randomized or only associational. experiment-simulate shows, on SIMULATED data, how often each label occurs. See growth experiments.

The experiment example contains illustrative post IDs that must be replaced with actual published posts and recorded native snapshots. Its readout requires matched ages and a predeclared hypothesis, metric, denominator, minimum pairs and guardrail; it never claims randomized causal proof. No proprietary ranking algorithm or guaranteed view count is assumed.

Claude operating methodology: identity/voice → audience and source research → evidence-bound campaign → truthful hooks and platform-native drafts → user review → approved outbox → native analytics → next test. The optimized router skill and on-demand references are in .claude/skills/reachweaver/; creative/evidence/experiment guides are progressively loaded rather than dumping 106 skills into the agent context. See competitive audit and agent evaluation cases.

Media asset checks (offline, Issue #18 slice)

reachweaver media-check media/manifest.json --json reads real PNG/JPEG/GIF/WebP/MP4/MOV headers. It checks dimensions, duration, size, extension mismatch, MP4/MOV video and audio codec, average frame rate, audio sample rate and channels, moov-before-mdat, sourced per-platform limits (unknown limits are skipped with a warning, never invented), image alt text, SRT/WebVTT captions for video, license and provenance (rights_confirmed must be true), and English-first translations. ReachWeaver does not upload or generate media; generation tools are optional, user-configured and need no paid model. See media assets.

reachweaver media-storyboard --campaign CAMPAIGN.json --out media/storyboard turns one campaign brief (the campaign contract or a signal-brief scaffold) into a vertical-video storyboard (JSON and Markdown), an SRT subtitle skeleton with valid timing, a media-check manifest with a LinkedIn image draft and the vertical video, and an image spec per platform. Alt text, captions, subtitle text and scene copy are {{write: ...}} placeholders and rights_confirmed starts false, so media-check blocks until a person writes them. It also flags text or focal boxes that a centered crop or a platform’s UI overlay would hide. The built-in TikTok, Instagram Reels and YouTube Shorts safe zones come from each platform’s first-party paid-ad creative specs (accessed 2026-10-09); organic overlays may differ, and every check says so. Supply other sourced numbers with --safe-zones. media-check also compares translated subtitle tracks cue by cue with the source track (cue count and timing, preserved terms, numbers and URLs).

First-party timing-outcome analysis (offline, observational)

timing-evaluate now compares already published posts in one configured account using the operator’s existing source-labeled native snapshot imports. It matches post ages and computes median rates by weekday/weekend six-hour audience-local dayparts. Missing values are excluded; an exploratory candidate needs observations in at least two time groups. It never fetches private analytics, changes a queue, guarantees reach or connects/publishes to a social platform.

reachweaver timing-evaluate --account my-bluesky --timezone Europe/Rome \
  --source native-export --metric clicks --denominator impressions \
  --age-hours 168 --tolerance-hours 24 --min-posts-per-slot 3 [--stratify format|campaign]

--stratify format (text/image/video, from the post’s stored attachment) or --stratify campaign adds the same readout per stratum. A stratum with too little data says insufficient_data, names no winner and points to calendar-plan’s generic baseline slots instead (cold_start); the pooled result does the same when it has too few comparable slots.

baseline-eval measures improvement against the non-personalized baseline: over a required --since/--until period it compares posts at your own --slot times (or in a --campaign-arm variant campaign) with posts at calendar-plan’s generic slot. Each comparison gets a seeded percentile bootstrap interval and a permutation p-value, and Holm–Bonferroni corrects across every comparison in the run, including saved experiment-readout files. Results are labeled explore or exploit and split into correlational and (owner-declared) randomized. Too few posts gives insufficient_data, never 0. Readout labels and the strict sign-test gate are unchanged.

reachweaver baseline-eval --account my-bluesky --timezone Europe/Rome \
  --source native-export --metric clicks --denominator impressions \
  --since 2026-09-01 --until 2026-10-31 --slot exploit:Tue@09:00 --slot explore:Sat@11:00

See timing evaluation and limits and calendar planner. native-export is a user-provided provenance label, not verification that analytics really came from the platform. Use experimental human-reviewed slots; the planner is not automatically retrained or scheduled by this result.

Social algorithms and best-time scheduling research (2026-10-08)

The Claude Code skill now includes platform algorithm/discovery guidance, an evidence-aware scheduling method, and a dated native-scheduling comparison. These distinguish provider scheduling UIs from authorized publishing APIs and the existing local outbox, including a platform-by-platform status matrix, audience-local timing experiments and a reviewable calendar schema.

Not implemented: automatic discovery of optimal hours from personal analytics, private platform analytics acquisition (only public Bluesky/Mastodon counters via metrics-fetch), a background scheduler, or platform-wide native/API scheduling. Time windows are observational starting points; not guaranteed visibility. ReachWeaver CLI remains approval-first and dry-run by default.

Safe batch calendar planning (experimental)

ReachWeaver now has a working offline calendar planner: after you write and individually approve posts, it can propose audience-local/UTC publication slots with low-confidence industry priors or your own explicitly supplied windows, avoid blackouts and collisions, and provide a reviewable JSON schedule. Only the existing Bluesky/Mastodon/Telegram/LinkedIn-member text-only adapters can accept --apply for local SQLite queueing, never immediate network publication. LinkedIn organization pages, YouTube, TikTok, Meta and other platforms stay native/manual scheduling handoffs until an enabled publishing adapter (the mock-tested ones above stay registry-disabled until a live receipt) and an account connection exist.

# Replace the dates with a future publication window.
reachweaver calendar-plan --account my-bluesky --timezone Europe/Rome --start 2026-10-15 --days 14
# After reviewing exact posts + approving each post, queue due times locally:
reachweaver calendar-plan --account my-bluesky --timezone Europe/Rome --start 2026-10-15 --days 14 --apply
reachweaver publish    # DRY RUN: no post sent, no background worker created

For platforms without a useful population-time prior (including Mastodon, Telegram and YouTube), provide account-selected slots such as --slot 'Tue@10:00' --slot 'Thu@16:00'. See calendar planner quickstart and algorithm/scheduling evidence. This feature does not learn an optimal time from connected analytics, automatically obtain credentials, run a daemon or guarantee visibility.

To plan from your own audience data, copy the hour-of-week “most active times” numbers from a native analytics dashboard you can see into ReachWeaver’s audience-activity CSV and pass --audience-activity FILE. The CSV is ReachWeaver’s own interchange format: no real platform export has been verified against it, and examples/audience-activity.example.csv is synthetic. The top activity hours become candidate slots. Every proposed entry carries provenance (first_party_audience_activity, the export’s source and as_of), and the output warns when as_of is older than 90 days. Hours you did not export stay missing instead of being treated as zero. Without the flag the output is unchanged. The planner stays advisory: it never approves or schedules anything on its own.

Guarded scheduled delivery and queue diagnostics (Issue #11 slice)

A guarded live send now has conservative per-account rolling 24-hour limits (default 2 attempts) and a 60-minute minimum gap, both checked atomically in the SQLite outbox. Past uncertain attempts still count even after restart/reconciliation. These are ReachWeaver limits, not cost controls.

Provider rate limits (owner-approved 2026-10-08). When a provider answers 429, ReachWeaver records a per-account cooldown from its Retry-After (or Telegram retry_after, Mastodon X-RateLimit-Reset, LinkedIn’s daily reset, otherwise 15 minutes), clamped to 60 seconds–24 hours. Until it ends, that account’s other posts are SKIPPED before any request and stay approved. The post that got the 429 goes to needs_review and is never resent automatically; dead-letter labels it certain_not_sent, likely_not_created or ambiguous to guide the human check. Only the read-only identity check before the claim may be retried (twice at most, never Bluesky). health, queue-health and the dry run show active cooldowns. config/caps.json version 2 adds tighten-only per-provider budgets; version 1 files still work. See provider rate limits. Since 2026-10-09 (owner-approved), Postiz-backed accounts that declare the same postiz_workspace share one 429 cooldown, and Bluesky connect-check/publish refuse before any request once the local createSession counter reaches half of Bluesky’s documented 30 per 5 minutes or 300 per day (bluesky_create_session_budget_exhausted); see account onboarding on shared Postiz workspaces and the identity check. All of this is mock-tested; no real provider 429 has been observed. The new offline queue-health command reports overdue local schedules, stale sending, needs_review, pause state and account attempt totals without connecting to social platforms.

reachweaver queue-health
reachweaver publish --scheduled-only --max-lateness-minutes 180 # DRY RUN
# Only after the owner's supervised live pilot, authorization and host credential setup:
# REACHWEAVER_ENV=prod REACHWEAVER_ALLOW_LIVE=1 reachweaver publish --live --scheduled-only --max-lateness-minutes 180 --max-per-24h 2 --min-gap-minutes 60

reachweaver health gives one offline readiness view per configured account (adapter, registry age, credential variable names, LinkedIn version and authorization expiry, Postiz, queue, rolling budget, pause, live opt-in and posts needing re-approval) and exits 1 when anything is blocked.

Approvals are bound to the exact payload (issue #7). approve stores a SHA-256 of the account, platform, exact text, schedule time, source URL and campaign (binding tag reachweaver.storage.approval/1), plus any image attachments (approval/media-1), video record (approval/video-1) or reply target (#17). A post that changes after approval is never sent: it stays approved/scheduled and is BLOCKED before any network call. Upgrading: posts approved or scheduled before this release are not grandfathered. health lists them as approval_rebind_required; re-approve each one with reachweaver unapprove POST_ID, reachweaver show POST_ID, reachweaver approve POST_ID (then schedule again). See the upgrade note. The hash detects unsanctioned edits; it does not prove a human ran approve (threat model T7/T8). reachweaver dead-letter groups needs_review posts by sanitized error code with reconcile steps. An optional config/caps.json (see config/caps.example.json) sets per-account caps that can only tighten the --max-per-24h/--min-gap-minutes values. Built-in tighten-only pilot caps (caps.BUILTIN_PROVIDER_CAPS, owner decisions D19-4 and D21-5, 2026-10-09) apply with or without a caps file: Threads accounts at most 2 sends per rolling 24 hours, Facebook Page accounts at most 2 per 24 hours and at least 60 minutes apart (Postiz-backed ones included), so a looser flag or file value cannot exceed them. A local config/optout.json lists handles or domains that are never replied to, mentioned or tagged (opt-out list); draft and approve refuse such mentions, and publish --live withholds an already-approved post before claiming it. For revoked permissions and API version sunsets, follow the runbook.

An unattended worker must use --scheduled-only to skip approved-but-unscheduled posts and missed stale time windows. The worker runbook includes inert, disabled-by-default systemd service/timer examples. No actual worker or account has been deployed by this repository, and uncertain sends still require human reconciliation.

Optional Postiz backend (Issue #30 slice, mock-tested)

An account can opt in to "backend": "postiz" so an approved, due text post is sent through your own Postiz instance or Postiz Cloud (Bluesky, Mastodon, Telegram, Threads, LinkedIn, Facebook only). ReachWeaver stays the approval authority: same --live + REACHWEAVER_ALLOW_LIVE=1 + REACHWEAVER_ENV=prod, pause, budgets and durable claim. A read-only integration check runs before the claim, and the key comes only from {env_prefix}_POSTIZ_API_KEY. A Postiz acceptance is not a platform receipt: the post goes to needs_review until you check the native post and reconcile. Reddit and X are refused. show POST_ID previews the exact Postiz request offline. If Postiz is down, the post stays approved, so you can post that text yourself and then cancel it here. Postiz is AGPL-3.0, and ReachWeaver only calls its HTTP API. The licensing and credential review (2026-10-15) and a ready owner smoke-test procedure are in the doc. The smoke test has not been run: no Postiz account has been connected or live-tested. See docs/POSTIZ.md.

Ranked publishing opportunities from your own sources (Issue #9 slice)

sources-scan reads an opt-in list of GitHub releases, tags, merged PRs and labelled closed-as-completed issues, RSS/Atom feeds, website or Keep a Changelog changelogs, and Crossref DOI records. It ranks them as SourceSignals with evidence URLs, freshness, novelty, audience relevance and a transparent score breakdown. Bot PRs, trivial PRs, drafts, private sources, reposts and items you have already seen are held back. By default it reads only local fixtures. --online makes explicit, GET-only requests to public GitHub repositories (paginated, capped by max_pages), https feeds and changelogs (after a robots.txt check), and the Crossref API. Fetched text is untrusted data, and the command never drafts or publishes. A private GitHub repository is read only with three opt-ins (visibility and sensitivity both private, plus --allow-private). This was owner-approved on 2026-10-09 (proposal). Its signals are marked private and are never public evidence or public links until a person marks each evidence record publishable. With --sources CONFIG, draft (reply drafts too), media-attach, youtube-attach and approve also refuse (and check reports as BLOCKED) a link into a visibility: private repository of that config (private_source_link). Without --sources these commands do not know the private repositories and cannot detect such a link, so pass it whenever a private source is configured. signal-brief then turns one saved signal into a new campaign scaffold plus a .brief.md sidecar: evidence URLs with an untrusted excerpt, one {{write: ...}} placeholder variant per approved account with a live or manual path, and the per-account length limits. It generates no copy, and the scaffold stays BLOCKED until a person writes every placeholder and verifies every source. See docs/SOURCES.md.

reachweaver sources-scan --config examples/sources.reachweaver.json --top 5
reachweaver sources-scan --config examples/sources.reachweaver.json --json > scan.json
reachweaver signal-brief --scan scan.json --signal-id SIGNAL_ID --brand examples/brand.reachweaver.json \
  --accounts config/accounts.example.json --out private/campaign.json

Which social networks to prioritize for real attention

ReachWeaver now includes an evidence-dated channel choice framework and platform-specific growth playbooks. A larger October 2026 open-source ecosystem audit compares leading social/marketing skills with schedulers, licenses and safe integration decisions.

• For AI, Kubernetes, open-source and research: start with LinkedIn plus original YouTube demonstrations. Use Reddit and Hacker News for selective, genuinely human-written community activity and technical feedback, not autonomous distribution.
• For broader consumer attention: reuse original demonstration footage as Instagram Reels, YouTube Shorts and TikTok-native edits if video-production capacity exists; then consider Threads/X/Bluesky for written conversations.
• For retention: opt-in Telegram/Discord/newsletter; these are generally not substitutes for public discovery.
• Data caveat: global monthly users, registered members, weekly users and Shorts video plays are fundamentally different. No channel guarantees views; measure first-party qualified outcomes.

Example Claude Code request:

/reachweaver — Audit the distribution options for my public GitHub release. Compare high-volume reach with technical audience fit using official 2026 sources. Prioritize two channels, give channel-native proof-backed drafts, explain platform-specific restrictions, and design a 14-day measured experiment. Do not auto-publish.

DEV Community: build technical authority without spam

DEV authority and collaboration playbook is a dedicated operating guide for creating a credible profile, discovering engineering readers, human-written comments, one evidence-rich article per week, selecting up to four relevant tags, using DEV’s native editor scheduler, and measuring qualified GitHub/collaboration outcomes. Use the bundled Claude Code specialist reference at .claude/skills/reachweaver/references/dev-community.md and the portable human-author brief at .claude/skills/reachweaver/assets/dev-article-brief.md.

Critical limits: DEV prohibits AI-generated comments/replies (apart from basic grammar/translation/accessibility assistance) and restricts AI-generated promotional/clout/backlink articles; read official rules. DEV has native article scheduling, but ReachWeaver has no DEV adapter, scheduled publishing receipt, account connection or native analytics fetch. For this channel, prepare an evidence inventory, author-edited brief and manual scheduling handoff—not fabricated automation.

Claude Code example:

/reachweaver — Audit my public DEV profile and my three best engineering posts. Recommend prioritized, evidence-checked profile improvements; a 30-day human-owned technical writing and community participation plan; a DEV native-scheduling checklist; and first-party metrics for GitHub contributors and real collaborations. Do not generate DEV comments or automatically publish.

Local dashboard and friendlier terminal output (experimental)

reachweaver ui serves a read-only dashboard of your local state on 127.0.0.1: what needs a human, the queue, each post’s exact text with a length meter and approval state, accounts and the offline channel research. It cannot approve, schedule, pause or publish. reachweaver start prints the exact next command for your current state, reachweaver completion bash|zsh prints a Tab-completion script, and at a terminal show prints a readable review sheet. See the dashboard guide.