signals-context-engineering
Inspect Signals across package boundaries or recover noisy command output with a bounded source map; skip narrow settled edits.
- License 许可
- MIT
- 最近更新
- 2026-10-24
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- signals-context-engineering
- description
- Inspect Signals across package boundaries or recover noisy command output with a bounded source map; skip narrow settled edits.
03 SKILL.md 正文
State the question and evidence needed. Start with rg --files, targeted searches and
narrow reads. For ownership questions spanning packages, run npm run context:pack
and inspect .meta/context/local/signals.xml. It is a disposable secret-scanned map
with a 30,000-token ceiling; reopen exact source before editing or asserting behavior.
It excludes private state, fixture data and historical reports.
When output is compressed, recover omitted detail from its raw_ref before rerunning
work. CCA recovery can be normalized/redacted and does not promise original bytes.
The optional host hook requires trust/activation; configuration does not establish
use. Keep generated output and receipts ignored and local.
Use focused references rather than loading every guide. New context tooling needs an observed navigation, fidelity or cost gap that searches, the map and raw recovery cannot solve. Compare critical-fact retention, reversibility, privacy, maintenance and real task outcomes; upstream savings are claims, not local results.
04 同仓库更多 Skills
signal-search
Discover reusable knowledge from live public sources using the invoking AI and its permitted tools; return evidence-bound findings and optional local saves without recursive AI execution.
repo-squash
Reconcile accumulated implementation, documentation, guidance and local artifacts into one small current product; use for repo cleanup or semantic squashing, not Git history rewriting.
signals-context-engineering
Inspect Signals across package boundaries or recover noisy command output with a bounded source map; skip narrow settled edits.
signal-search
Discover reusable knowledge from live public sources using the invoking AI and its permitted tools; return evidence-bound findings and optional local saves without recursive AI execution.
repo-squash
Reconcile accumulated code, documentation, guidance and local artifacts into one small current product.