القائمة
Use when akari needs to proactively scan for new GenAI developments — model releases, capability changes, relevant research — rather than waiting for human input
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Use when a file or module is too large, responsibilities are tangled, or a cross-cutting redesign is needed
Use when literature notes exist and need citation verification, or before publishing any artifact that cites literature
Use at the end of a work session to embed learnings into conventions. Simplified for models with lower reasoning capacity.
Use at the end of a work session, or when accumulated findings need to be embedded into conventions, skills, or patterns
Use when the Slack bot needs to handle an operational action like launching an experiment, querying status, or processing approvals
Use when a plan, finding, or design needs adversarial review before committing to it
| name | horizon-scan |
| description | Use when akari needs to proactively scan for new GenAI developments — model releases, capability changes, relevant research — rather than waiting for human input |
| complexity | medium |
| model-minimum | glm-5 |
| disable-model-invocation | false |
| allowed-tools | ["Read","Write","Grep","Glob","WebSearch","WebFetch","Bash(git *)"] |
Proactively scan external sources for GenAI developments relevant to akari's active projects. This skill replaces human knowledge injection for routine developments (new model releases, benchmark results, capability announcements, relevant papers).
If a focus area is provided (e.g., "3D generation models", "LLM judge evaluation"), narrow the scan to that domain. Otherwise, scan broadly across all active project concerns.
Read the following to understand what akari currently cares about:
ls projects/ → read each active project's README for Mission, Open questions, and recent log entries..scheduler/skill-reports/horizon-scan-*.md for previous scans to avoid redundant coverage.From this, produce a scan agenda: 3-5 specific topics to search for, each tied to a project need or open question. Example:
Scan agenda:
1. New model releases since last scan (if you track models)
2. 3D generation quality benchmarks or evaluations (sample-project open questions)
3. LLM agent architecture papers (akari design-patterns paper related work)
4. Multi-modal evaluation methods (sample-project judge methodology)
For each topic in the scan agenda, run 2-3 WebSearch queries. Use time-bounded queries where possible (e.g., include the current month/year to find recent results).
Search strategy:
Collect all results. Aim for 10-20 raw results across all topics.
For each search result, classify:
Verification (mandatory for all actionable and informative items):
This follows the same URL verification discipline as ADR 0019 and /lit-review. Parametric memory is not evidence.
For each verified finding, determine where it belongs:
→ If you maintain a model registry in your own system, update it:
→ MANDATORY: Create a task in the relevant project's TASKS.md:
- [ ] Evaluate [model name] against known capability baselines
Why: Horizon-scan detected new release. [1-sentence summary of claimed improvements].
Done when: Model evaluated with a documented protocol; decision recorded.
Priority: medium
Source: horizon-scan YYYY-MM-DD, [URL]
Every model release that affects an active project requires a follow-up task. If the release is irrelevant to all active projects, note this explicitly in the scan report.
→ If load-bearing (directly addresses an open question), create a literature note following the /lit-review schema in the relevant project's literature/ directory. Mark Verified: YYYY-MM-DD.
→ If load-bearing and actionable, also create a task in the relevant project's TASKS.md (follow /lit-review Step 6 format).
→ If contextual (useful background), note in the scan report but do not create a full literature note unless the project has an active literature task.
→ Update the relevant project file (e.g., existing-data.md, README open questions, or a dedicated tools/APIs tracking file).
→ If the change unblocks a task, update the task's lifecycle tags.
→ If the change enables new work, create a task in the relevant project's TASKS.md.
→ Record in the scan report only. Do not create project artifacts for developments that don't clearly affect an active project.
Write a structured report to .scheduler/skill-reports/horizon-scan-YYYY-MM-DD.md:
# Horizon Scan: YYYY-MM-DD
Scope: [focus area or "broad"]
Topics scanned: [list from scan agenda]
Sources checked: [count of URLs fetched and verified]
## Actionable findings
### [Finding title]
- **What**: [1-2 sentence summary]
- **Source**: [verified URL]
- **Affects**: [project name, capability, or open question]
- **Action taken**: [what was updated — registry, task created, literature note, etc.]
### [Finding title]
...
## Informative findings
- [Brief summary with source URL] — relevant to [project/question]
- ...
## No-signal topics
- [Topic]: No new developments found since [last scan date or "initial scan"]
- ...
## Gaps
- [Topic where verification failed or sources were unavailable]
- ...
After recording all findings, assess whether any finding changes priorities:
Follow docs/sops/commit-workflow.md. Commit message: horizon-scan: [date] — [N] actionable, [M] informative findings