agent/claude/issues/000-overview.md

feat: Replace shell scripts with core CLI commands

Summary

This tracking issue covers the migration from shell scripts to native core CLI commands in core-agent. Every shell script should be replaced by a core command, making the plugin purely configuration (JSON, MD) + calls to core.

Philosophy

• Every agent using core tests the framework
• Missing features get raised as issues
• Code as if functionality exists (TDD approach)
• The CLI becomes battle-tested and bulletproof

Issues

Claude Code Hooks (core ai)

Issue	Command	Replaces
#001	core ai session	pre-compact.sh, session-start.sh, suggest-compact.sh
#002	core ai context	capture-context.sh, extract-actionables.sh
#003	core ai hook	prefer-core.sh, block-docs.sh, post-commit-check.sh

Quality Assurance (core qa)

Issue	Command	Replaces
#004	core qa debug	check-debug.sh

Data Collection (core collect)

Issue	Command	Replaces
#005	core collect github	github-history/collect.sh
#006	core collect bitcointalk	bitcointalk/collect.sh
#007	core collect market	coinmarketcap/*.sh
#008	core collect papers	ledger-papers/discover.sh, collect-whitepaper.sh
#009	core collect excavate	project-archaeology/excavate.sh
#010	core collect process	job-collector/process.sh
#011	core collect dispatch	collection/dispatch.sh

Target State

After all issues are implemented:

claude/
├── hooks/
│   └── hooks.json          # Calls core ai hook validate-*
├── commands/
│   └── remember.md         # Calls core ai context add
├── collection/
│   └── hooks.json          # Calls core collect dispatch
└── skills/
    └── */
        └── SKILL.md        # Documentation only, calls core collect *

No shell scripts. Just JSON config + markdown docs + core CLI calls.

Hook Improvements (feedback cycle)

Issue	Feature	Purpose
#012	Test output filtering	Reduce noise, show only failures
#013	Stop verification	Verify work complete before stopping
#014	Auto-test on edit	Run tests async after code changes
#015	Session context	Inject git/issues/CI context on start
#016	Silent formatting	Auto-format without output noise
#017	Expose/hide policy	Define what to show vs suppress

Implementation Order

1. Phase 1: core ai session + core ai context (enables hooks to work)
2. Phase 2: core ai hook + core qa debug (safety + quality)
3. Phase 3: Hook improvements (012-017) - better feedback cycle
4. Phase 4: core collect github + core collect bitcointalk (most used)
5. Phase 5: Remaining collection commands