Remove foreign skill projection from aoa-techniques

[8Dionysus]

Summary

• remove the foreign .agents/skills/ cache: 25 shared bundles and 212 copied paths, including six helper scripts
• establish the accepted no-home boundary in AOA-TECH-D-0069: no empty skills/ port and no repo projection until a repo-specific capability proves admission
• update AGENTS routes, semantic/script topology coverage, decision indexes, doc read models, and the repo-local KAG family

Manual evidence first

• baseline prompt inspection exposed 10 repo-local copied bundles
• post-removal prompt inspection exposes no aoa-techniques repo skill; shared AoA skills still arrive from the user profile
• a positive clean-agent trial found AOA-T-0002, returned to its authored TECHNIQUE.md, and preserved the negative owner boundary without a local skill
• a negative clean-agent trial rejected a proposed broad search/select/adapt/validate/retry/publish skill and routed proposal pressure through Method-growth and AOA-T-0102

No raw session trace or session identifier was promoted into owner truth.

Validation

• python scripts/ci_gate.py --mode source-fast with pinned aoa-kag and current aoa-stats owner roots
• python scripts/ci_gate.py --mode generated
• python scripts/release_check.py
• canonical repo-local KAG full and incremental parity plus family validation
• live codex debug prompt-input inspection after removal

The release pass covered 249 root tests, 35 part-local tests, all 15 mechanics test directories, 107 technique contracts, generated parity, public hygiene, Spark, stats, and KAG export admission.

Boundary and residual risk

This does not claim that skills are globally unnecessary. It records that no current aoa-techniques-owned candidate has proved a distinct trigger, ABI, composition value, failure boundary, and repeatable benefit against no-skill and coexistence baselines. That boundary is reversible through future manual evidence.