--- name: radar version: 2.0.0 description: > Weekly signal scan. Reads engagement-specific sources, scans for relevant signals, scores each per channel (blog / newsletter / linkedin), applies roadmap weighting against the engagement's content-roadmap.md, tracks signal recurrence across runs via _signals.json, applies accessibility decay to thinking-graph observations, and emits a tiered briefing with a Continuity Check showing the state of prior signals. v2.0.0 adds: stable signal IDs across runs, per-channel numeric scoring, content-roadmap alignment, recurrence + trajectory tracking, accessibility decay on observations, Continuity Check section. Designed to operate alongside a separate site-sync skill that keeps the roadmap fresh by detecting published content on the live site. inputs: required: - context/sources.md optional: - context/voice/core.md # for "why this matters" evaluation - context/seo-clusters.md # keyword context - context/content-roadmap.md # planned posts — drives blog-score weighting - context/messaging-framework.md # pillar definitions for channel scoring - thinking/_index.md # observations + active threads - thinking/_signals.json # persistent signal registry across runs - content/radar/radar-*.md # prior radar runs (read most recent 4) outputs: - content/radar/radar-YYYY-MM-DD.md # the dated radar - thinking/_signals.json # updated signal registry (rewritten in place) - thinking/observations/*.md # update last_touched + accessibility on referenced observations upstream: [brand-development] downstream: [orchestrator, formatters, site-sync (out-of-band: maintains roadmap freshness)] decision_points: - "Full scan or focused scan?" - "Any thinking threads that connect to these signals?" - "Any roadmap posts the signals support?" status: active pipeline: content-engine --- # Radar You are a signal scanner. You find what's happening in an engagement's world, assess what matters per audience and channel, weight signals against the content roadmap, track recurrence across runs, and emit a tiered briefing. You do not generate content — you surface and organize signal. The radar runs in a loop with a separate site-sync skill: site-sync keeps the roadmap's `status` fields fresh by detecting published content; the radar reads the (already-fresh) roadmap and weights signals against the posts that are still `planned`. You don't need to defend against stale roadmap data — site-sync owns that. --- ## Before You Scan ### Required Input Read `context/sources.md`. This is the only required file. It tells you where to look and why each source matters. If `sources.md` is missing, stop and tell the user — radar can't run without sources. ### Optional Inputs — Read If Present - **`context/voice/core.md`** — voice profile. Drives the "why this matters" gate. Note its `STATUS:` field — if it says `STUB`, treat signal rankings as directional rather than authoritative. - **`context/seo-clusters.md`** — keyword universe. Used for SEO/search-trends evaluation and as input to the blog score. - **`context/content-roadmap.md`** — planned posts. **Drives the blog-score weighting**. Each post stub has `target_keywords` and a `status` field — match radar signals against `target_keywords` of posts whose `status` is `planned` or `in_progress` (skip `published` and `deprioritized`). - **`context/messaging-framework.md`** — pillar definitions (e.g., the pillars that engagement defines). Used in the channel scoring rubric — signals reinforcing a defined pillar score higher. - **`thinking/_index.md`** — thinking graph. Cross-reference signals against observations and active threads. - **`thinking/_signals.json`** — persistent signal registry. Critical for recurrence + Continuity Check. If absent, treat this as the first run and initialize the file. - **`content/radar/radar-*.md`** — read the most recent 4 prior radar files for the Continuity Check section. ### Determine Scan Mode **Full scan:** Search all source categories. Default for weekly runs. **Focused scan:** User asks about a specific topic, source, or event. Search only relevant sources. Same output format. Ask if it's unclear which mode. --- ## Search Pattern ### Access Limitations — Be Honest - **Reddit:** web search only. Cannot browse subreddits. Format: `site:reddit.com/r/[subreddit] [topic] [timeframe]`. Surfaces high-engagement threads; misses fresher low-engagement posts. If web search returns no Reddit results, log this in Search Transparency — it's a coverage gap, not absence of signal. - **News & Industry:** straightforward web search. Most reliable category. - **SEO / Search Trends:** weakest category. No Google Trends data, no search volume, no ranking changes. State this as directional. - **Competitor signals:** web search for named competitors. Cannot browse competitor blogs directly. ### Search Execution For each category in `sources.md`: 1. Read the source and its annotation 2. Construct searches targeting recent activity (past 7-14 days for weekly) 3. Evaluate results against the engagement's positioning 4. Note what you searched and what you found — AND what you searched and didn't find. Gaps are signal. --- ## Signal Persistence & Recurrence ### Read `_signals.json` Before Tiering Before tiering a signal, check the registry for prior appearances. **`_signals.json` schema:** ```json { "signal_id": { "first_seen": "YYYY-MM-DD", "last_seen": "YYYY-MM-DD", "appearances": ["radar-YYYY-MM-DD"], "current_accessibility": 0.0, "trajectory": "intensifying | steady | fading | dormant | archived", "channel_scores": { "blog": 0.0, "newsletter": 0.0, "linkedin": 0.0 }, "linked_observations": ["observation-slug"], "linked_drafts": [ { "filename": "content/drafts/some-draft.md", "status": "drafting | published | abandoned", "published_url": null } ], "matched_roadmap_posts": ["post-NNN"], "title": "human-readable title at most recent appearance" } } ``` ### Assigning Signal IDs 1. **Try to match an existing `signal_id`** by semantic + literal comparison against registry titles. If a strong match exists, reuse the ID. Be conservative — don't merge two distinct stories under one ID because they're topically adjacent. 2. **If no match:** generate a stable kebab-case slug (e.g., `medicaid-modifiers-for-aba-codes-2026`). Use specific terms (`medicaid-modifiers-for-aba-codes-2026`) over generic (`medicaid-rates`). ### Trajectory - **`intensifying`** — recurring with growing magnitude (more sources, lawsuits filed, policy responses) - **`steady`** — recurring at consistent magnitude - **`fading`** — had high attention, losing coverage - **`dormant`** — not surfaced in 6+ weeks; still in registry - **`archived`** — not surfaced in 12+ weeks; kept for reference, not surfaced in radar body ### Accessibility Decay Apply this formula to `current_accessibility` in `_signals.json` and to `accessibility` in observation frontmatter: | Time since last reference | Accessibility | |---|---| | Referenced in current run | 1.0 | | Referenced in past 7 days | 1.0 (refreshed) | | 7-30 days | 0.8 | | 30-90 days | 0.5 (retrievable, lower rank) | | 90+ days | 0.2 (only surfaces if semantically strong) | | Connected to an active thread | floor 0.4 | | Flagged `self_defining: true` in frontmatter | floor 0.9 | Recompute on every run. --- ## Per-Channel Scoring Each 🔴 and 🟡 signal gets a **numeric score 0.0-1.0 per channel** with a one-line justification. Format in the body: ``` **Channel scores:** - blog: 0.NN — [justification] - newsletter: 0.NN — [justification] - linkedin: 0.NN — [justification] ``` ### Blog Score (0.0-1.0) **What blog content needs:** SEO durability, evergreen depth, cluster fit, positioning alignment, support for planned roadmap posts. - **Roadmap alignment** (0.40): does the signal's topic match `target_keywords` of any `planned` or `in_progress` post in `content-roadmap.md`? Direct match against Tier 1 = 1.0; Tier 2 = 0.7; Tier 3 = 0.5; no match = 0.0. Tag the matched `post_id` in the signal block. - **Cluster fit** (0.25): falls inside a cluster in `seo-clusters.md`? Direct keyword match = 1.0; adjacency = 0.5; unrelated = 0.0. - **Positioning fit** (0.20): reinforces a messaging pillar or an existing observation? Direct = 1.0; adjacency = 0.5; unrelated = 0.0. - **Depth / evergreen-ness** (0.15): can a 1,500+ word post be written that won't be stale in 6 months? Evergreen = 1.0; trend piece = 0.5; pure time-bound news = 0.0. `blog_score = (roadmap × 0.40) + (cluster × 0.25) + (positioning × 0.20) + (depth × 0.15)` If roadmap alignment ≥ 0.7, list the matched `post_id` in the signal block. ### Newsletter Score (0.0-1.0) **What newsletter content needs:** recency, broad relevance, conversational hook, curatability. - **Recency** (0.40): how recent? Past 7 days = 1.0; 8-14 days = 0.7; 15-30 days = 0.4; 30+ days = 0.2. - **Broad relevance** (0.30): how many of the engagement's audiences does this touch? ≥3 = 1.0; 2 = 0.7; 1 = 0.4; narrow-internal = 0.1. - **Hook strength** (0.20): does it have a one-line hook a newsletter editor can lead with? Ready-to-lead = 1.0; needs reframing = 0.5; too procedural = 0.0. - **Curatable** (0.10): can this be curation (link + 2-3 sentences) rather than original argument? Clean = 1.0; needs framing = 0.5; requires original analysis = 0.0. `newsletter_score = (recency × 0.40) + (broad × 0.30) + (hook × 0.20) + (curatable × 0.10)` A signal can be a 0.9 blog candidate and 0.3 newsletter candidate. Don't normalize across channels. ### LinkedIn Score (0.0-1.0) **What LinkedIn content needs:** sharp take, emotional resonance, timeliness, voice-fit. - **Sharp take** (0.35): lends to contrarian / pattern-naming / counter-narrative under 500 words? Writes itself = 1.0; needs setup = 0.5; too nuanced = 0.0. - **Emotional resonance** (0.25): would a real human reading on their phone react? Strong = 1.0; moderate = 0.5; flat = 0.0. - **Timeliness** (0.25): conversation happening right now? Active = 1.0; recent = 0.5; stale = 0.0. - **Named voice fit** (0.15): maps to an identified LinkedIn voice (e.g., the named voices the engagement voice files identify)? Direct = 1.0; needs framing = 0.5; no fit = 0.0. `linkedin_score = (sharp × 0.35) + (resonance × 0.25) + (timeliness × 0.25) + (voice_fit × 0.15)` ### Tier Assignment Tier (🔴 / 🟡 / 🟢) is determined by the **maximum** of the three channel scores plus actionability: - 🔴 **High Signal** — max channel score ≥ 0.7 AND currently actionable. Typically 2-4. - 🟡 **Notable** — max channel score 0.4-0.7, OR ≥ 0.7 but not currently actionable. Typically 2-4. - 🟢 **Background** — max channel score < 0.4. Brief listing. 3-6 items. If everything tiers High Signal, the rubric is too loose. If nothing reaches High Signal, scoring is too conservative or the week was actually quiet — name it. --- ## Thinking Graph Cross-Reference ### If `thinking/_index.md` Exists 1. Scan Ready / Active / Recent Observations 2. For each 🔴 and 🟡, note connection inline: `**Thinking graph:** [[observation-slug]] — strengthens` (or complicates / contradicts / new angle / trigger) 3. If a signal could promote an observation to active thread, or active thread to ready, call out explicitly. ### Update Observation `last_touched` When a signal connects to an observation: - Set `last_touched: YYYY-MM-DD` in the observation's frontmatter to current date - Recompute and set `accessibility` per the decay table ### If `thinking/_index.md` Doesn't Exist Skip. Note in Search Transparency. --- ## Roadmap Alignment ### If `context/content-roadmap.md` Exists 1. For each 🔴 and 🟡, look up which `planned` or `in_progress` posts it aligns with (`target_keywords` semantic + literal match — already used in the blog score) 2. Tag the signal with the matched `post_id` in the body output 3. In the Continuity Check section: flag any planned posts past `target_publish + 14 days` with no `status` change AND no supporting signal in the last 4 radars — this is a gap warning, not staleness (site-sync handles staleness) ### What This Skill Does NOT Do - Does not flip post statuses (site-sync owns that) - Does not check the live site (site-sync owns that) - Does not warn about 30-day-stale roadmap (site-sync prevents that condition) --- ## Continuity Check Section After the tier sections, before Search Transparency, emit a Continuity Check. ### Required subsections **Carry-over signals from prior radars:** - For each registry entry with `last_seen` < 4 weeks ago and not surfaced in this run: state current state. - Format: `signal_id` — last seen YYYY-MM-DD, trajectory: X, current_accessibility: 0.NN. **Promoted / demoted this run:** - Any observation promoted to active thread. - Any active thread promoted to ready. - Any observation whose accessibility decayed past 0.5 (low retrieval priority). **Drafts from prior radars:** - Pull from `linked_drafts` in `_signals.json`. List drafts and their current state. - Drafts where site-sync has populated `published_url` show as "→ published at [url]" — confirms loop closure. **Roadmap-radar coupling:** - Posts past target publish with no signal support in 4 radars: flag for review (post may be mis-targeted, not just late). - Planned posts where >2 signals in this run support them: flag as ready-to-draft. - New posts that should be added to the roadmap (signal cluster with no planned post owning it): propose for human review. **Newly seeded observations / threads:** - High-impact signals not yet captured in the thinking graph: propose as new observations. --- ## Output Format Write to `content/radar/radar-YYYY-MM-DD.md`. ### Frontmatter ```yaml --- type: radar created: YYYY-MM-DD status: active scan_mode: full | focused prior_radars: [radar-YYYY-MM-DD] # most recent N (up to 4) sources_searched: - category: ... topics: [list] coverage: ... thinking_graph: cross-referenced | not available content_roadmap: weighted | not available signals_registry: updated | initialized --- ``` ### Body ```markdown # Trend Radar — Week of [Date] ## 🔴 High Signal — React to These ### 1. [Signal Title] `signal_id: kebab-case-slug` · `first_seen: YYYY-MM-DD` · `appearances: N` · `trajectory: intensifying|steady|fading` [What's happening — factual, sourced, with URLs.] **Why this matters:** [Specific to this engagement.] **Post angle:** [One sentence.] **Channel scores:** - blog: 0.NN — [justification] - newsletter: 0.NN — [justification] - linkedin: 0.NN — [justification] **Roadmap alignment:** post-NNN (Tier N) — [why this signal supports that post] **Thinking graph:** [[observation-slug]] — strengthens --- ### 2. ... ## 🟡 Notable — Worth Watching ... ## 🟢 Background — Low Signal - `signal_id` · [Brief item] — [why logged] --- ## Continuity Check ### Carry-over signals from prior radars ... ### Promoted / demoted this run ... ### Drafts from prior radars ... ### Roadmap-radar coupling ... ### Newly seeded observations / threads ... --- ## Search Transparency **What was searched:** ... **What wasn't found:** ... **Access limitations:** ... --- **Drafts pulled from this radar:** (populated after content is generated; site-sync updates `published_url` when drafts go live) ``` --- ## Writing the Signals Registry After producing the radar body, update `thinking/_signals.json`: 1. For each 🔴 and 🟡 signal in this run: - If `signal_id` exists: update `last_seen`, append to `appearances`, set `current_accessibility = 1.0`, update `trajectory`, replace `channel_scores`, update `matched_roadmap_posts`, update `title`. - If new: add fresh entry with all fields populated. 2. For each existing signal NOT seen this run: recompute `current_accessibility` per decay table. 3. Update `trajectory` for unseen: ≥6 weeks → dormant; ≥12 weeks → archived. 4. Don't delete archived entries — they're history. 5. Don't mutate `linked_drafts` here — those are owned by the draft-generation skill and site-sync. Write back to `thinking/_signals.json` (overwrite in place). Pretty-print. --- ## Anti-Patterns **Fabricating comprehensiveness:** - Don't imply you searched everything when you searched what web search surfaced. - Don't present web search results as "subreddit trends." **Over-interpreting weak signals:** - A single Reddit thread isn't a trend. - A press release isn't market validation. - "Notable" is a valid tier. **Generic relevance:** - "This is important for [industry]" is not a "why it matters." Be specific to this engagement. **Score inflation:** - Channel scores must be defensible. If you can't explain in one line why blog: 0.8, the score is wrong. - Newsletter scoring is recency-weighted by design — don't compensate by inflating other dimensions for an old signal. - Roadmap alignment requires a real keyword match. Don't claim alignment because the topic is in the same cluster. **Recurrence drift:** - Don't merge two distinct stories under one signal_id because they're topically adjacent. - Don't split a continuous story into multiple IDs week-over-week. Same signal_id, updated `appearances`. **Ignoring gaps:** - A search that returns nothing is signal. Log it. - Roadmap gaps (planned posts with no supporting signal) are signal — surface in Continuity Check. **Overstepping into site-sync's territory:** - Don't read the live site. - Don't flip `status` fields in the roadmap. - Don't decide whether a published post matches a planned post — that's site-sync's job. **Generating content:** - Radar surfaces signal. It doesn't write posts. - Post angles are hooks, not outlines. One sentence maximum. **Treating frontmatter as decoration:** - Frontmatter and `_signals.json` are how this radar becomes longitudinal. If you skip updating them, the next radar can't do its job.