domain-watch

name	domain-watch
description	Autonomous perception layer - scans KB for new notes matching domain watch configs, checks gap resonance with the Thinking Registry, probes external signals via web search, and auto-activates HIGH/MEDIUM signals into the Thinking Registry for the incubation loop.
automation	autonomous
schedule	0 7 * * *
allowed-tools	Read, Write, Grep, Glob, Bash, WebSearch
user-invocable	true
argument-hint	[domain-slug] (optional - scans all active domains if omitted)
metadata	{"version":"1.18","updated":"2026-07-18T00:00:00.000Z","changelog":["1.18: Step 6 refusal-≠-discard rule — a wire dated stale and refused can still carry base-rate color: if its dating reveals a prior instance of a class the topic priced as a single event, post the color (zero weight, channel-scoped) alongside the reject-tier guard (Run 837: the Apr-5-2026 Kuwait power/desalination item, refused for E4 leg-A, revealed the pre-era precedent is a ≥2-instance class bearing on the March-anchor channel citation)","1.17: Step 6 two dating-hazard refinements — (a) own-channel ≠ fresh: an institution's or provider's OWN channel can serve stale pages on generic queries (Run 836: fm.gov.om's official round-announcement page dated internally to May 2025; gcc-sg.org's March-2026 Extraordinary-statement page), so the v1.14 primary-pin grade applies only after the page itself is dated from internal anchors; (b) weekday anchors lose discriminating power when a prior instance of the same event class is exactly 7n days old (Run 836: the Jul-11-Saturday Araghchi-Busaidi artifact vs the live Jul-18-Saturday round) — on weekday collision, discriminate by principals and content, never weekday alone","1.16: Step 6 pin-completion criterion — a later scan completes a v1.15 candidate-primary pin by CONTENT-match (a readable own-channel post from the same provider stating the same figure, date, and decomposition in the provider's own words), NOT by status-id match; the readable post may carry a different id than the one named at candidate grade (name the supersession in the injection). Run 835: the Jul 16 8-print pinned via readable post 2078085624439898314, superseding the unread candidate id 2077331039458865604 named at Run 834","1.15: Step 6 candidate-primary grade — when the targeted verification query SURFACES the data provider's own channel post (right provider, right query, newer than consumed posts) but its TEXT cannot be read, grade the datum as multi-carriage + candidate-primary: above the v1.13 single-carriage floor, below the v1.14 read-primary pin; name the unread post (URL/status id) in the injection so a later scan can complete the pin (Run 834: the Jul 16 'crossings = 8' print stood on three carriages + an unreadable Kpler post)","1.14: Step 6 primary-pin upgrade path — a verification probe that re-surfaces the DATA PROVIDER'S OWN channel post (e.g. Kpler's or MarineTraffic's X post stating the figure, date, and basis in its own words) counts as a PRIMARY pin at full grade, and a second independent provider's own-channel post upgrades the datum further (two primary pins > any single wire carriage); complement of the v1.13 failed-re-pin single-carriage rule (Run 833: the quarantined '21 ships Tuesday' resolved at full grade on Kpler's own post + MarineTraffic's independent post, executing the basis-resolution order wire carriages alone could not settle)","1.13: Step 6 verification-probe grading — when a v1.7 verification probe FAILS to re-surface a candidate wire's primary source (returns only older vintages), the datum is neither refused nor accepted at full grade: carry it at reduced SINGLE-CARRIAGE grade, name the carriage outlet, and flag the grade in the injection so the consuming run weighs it (Run 832: the '11 vessels Wednesday' Bloomberg-derived figure stood on one Spokesman-Review carriage after the re-pin probe returned only June vintages)","1.12: Step 6 stale-wire consumption checks extend to LINEAGE — before treating a wire as a fresh arrival for an active topic, grep the predecessor/converged sibling thinking files too: a wire consumed in a predecessor (e.g. the Oman two-corridor draft, E-B in hormuz-toll-regime-institutionalization Run 802) can re-surface against the successor topic and masquerade as fresh; post the reject-tier guard to the ACTIVE topic naming where it was consumed (Run 830)","1.11: Step 6 stale-wire discipline sharpened — the search tool's own result SUMMARY is an untrusted dating source: it can assert a date the underlying wire's internal anchors refute (Run 829: summary claimed a 'July 16' Kpler print; the weekday anchor 'Wednesday' + the 14→13→9 series identified it as the consumed Jul 15 print). Date from the wire's internal anchors, never from the summarizer's attribution","1.10: Step 6 stale-wire dating discipline — search results for recurring event classes (talks rounds, strikes, counts) often surface undated articles from PRIOR instances of the same event class; before treating a wire as a fresh arrival, date it from internal anchors (weekday references, named context era, actors) and refute-or-corroborate via a v1.7 verification probe; a refuted stale wire gets posted to the topic as a reject-tier stale-artifact guard so subsequent runs don't re-trip on it (Run 828: prior-round 'Muscat cancelled' piece vs the standing Jul 18 round)","1.9: Round-counting discipline for accruing-window discriminators (e.g. an inherited D-table's Nth-round watch): when a fresh report differs from an already-consumed round only in magnitude or target detail and no distinct time window is establishable, do NOT advance the round counter on interpretation — carry it as an explicit revised-counts-vs-new-round ambiguity with conservative counting and flag it in the injection for the next run to argue (Run 824: Kuwait 6+33 vs consumed 4+21 within the same Jul 15 reporting day)","1.8: Step 8.5 pre-committed-successor guard — a converged topic's pre-authorized successor whose trigger (a named discriminator/event) has NOT yet fired is not an eligible floor candidate; seeding it early re-opens the converged topic. Floor elsewhere; when the discriminator fires it arrives as a HIGH signal → Step 8 spawn (Run 818)","1.7: Step 6 verification exception — a candidate pre-registered event or structural break for an ACTIVE topic justifies extra targeted searches to date/corroborate that one item before injecting (verify the arrival, not widen the sweep; Run 813)","1.6: Step 1 UTC discipline extended to find -newermt — it parses timestamps in container-local time, so a bare UTC scan time silently empties the changed-since window (bit Run 812); use git --since with explicit +00:00 offset, or find with an explicit offset","1.5: Step 3 LBS invocation fixed to the run_search.sh wrapper — calling search.py with bare python3 fails on the Trinity container (ModuleNotFoundError: faiss; deps live in the wrapper's environment)","1.4: Step 8 injection-not-spawn rule — a HIGH/MEDIUM signal answering an ACTIVE topic's named watch/pre-registered event is appended to that thinking file as a dated evidence injection (with divergence flags), never spawned as a duplicate topic (pattern: Runs 806/807/809)","1.3: Step 1 UTC discipline — use date -u and explicit-offset git --since; the Trinity container clock is non-UTC (observed UTC-4), which mislabels scan times and silently shifts change windows","1.2: Starvation Floor duplicate guard — before ranking a PIR as uncovered, check the ENTIRE Thinking Registry (converged/crystallized/retired, not just active) + Thinking/ filenames for the candidate keywords","1.1: Thinking Registry write discipline — topic activation adds a table ROW only; never write scan narratives or frontmatter fields (e.g. last_domain_watch) into THINKING-REGISTRY.md; scan results belong in the Watch Log and domain configs","1.0: Initial version"]}

Domain Watch

ℹ️ First, set expectations: before anything else, print one short line with this skill's version and its most recent change — the top entry of metadata.changelog above — e.g. domain-watch vX.Y — recent: <summary>. Then proceed.

Autonomous perception layer. Each scheduled run scans configured domains for new KB notes, checks gap resonance with the Thinking Registry, probes external signals via web search, and auto-activates HIGH/MEDIUM proposals directly into the Thinking Registry. LOW/NONE signals are logged only.

Design principle: Perceive, score, activate. Human reviews the Watch Log post-run.

State Dependencies

Source	Location	Read	Write	Description
Watching Registry	Brain/05-Meta/Watching/WATCHING-REGISTRY.md	✓	✓	Active domain list + last scan dates
Domain Configs	Brain/05-Meta/Watching/[domain-slug].md	✓	✓	Per-domain PIRs, queries, baseline counts
Watch Log	Brain/05-Meta/Watching/WATCH-LOG.md	✓	✓	Append-only proposals and scan history
Thinking Registry	Brain/05-Meta/Thinking/THINKING-REGISTRY.md	✓		Gap resonance check (read-only)
Local Brain Search	resources/local-brain-search/	✓		Semantic search for KB notes
Session Changelogs	Brain/05-Meta/Changelogs/		✓	Run log

Prerequisites

• Watching Registry exists at Brain/05-Meta/Watching/WATCHING-REGISTRY.md
• At least one active domain seeded (see Seeding section below)
• Local Brain Search index up-to-date

---

Process

Step 1: Get Date and Load Registry

date -u '+%Y-%m-%d %H:%M UTC'

Always use date -u: the Trinity container clock runs in a non-UTC local zone (observed UTC-4), so a bare mislabels the scan time and a given a UTC timestamp silently shifts the change window by hours. When computing "changed since last scan" with git, pass with an explicit offset (e.g. ). The same trap applies to - it parses the timestamp in CONTAINER-LOCAL time, so a bare UTC scan time silently shifts the window ~4h forward and returns an empty/short list (bit Run 812). Prefer with the explicit offset for the changed-since check; if using find, append an explicit offset (, GNU find).