feat(plugin): add Claude Code plugin with guided skills, agents, and hooks

[rsmets]

Summary

Adds a local Claude Code plugin (abca-plugin/) that gives developers guided workflows for setting up, deploying, operating, and troubleshooting the ABCA platform, directly from their IDE or terminal.

In response to Issue #17.

• 7 skills: setup, deploy, onboard-repo, submit-task, troubleshoot, abca-status, abca-submit
• 2 agents: cdk-expert (infrastructure work), agent-debugger (task failure investigation)
• 1 hook: SessionStart injects project context into every Claude Code session

How to use

claude --plugin-dir abca-plugin

Area

• docs — guides or design sources (docs/guides/, docs/design/)
• tooling — root mise.toml, scripts, CI workflows

Changes

• Created abca-plugin/ directory with plugin.json manifest
• Added 7 skills as SKILL.md files with trigger descriptions and step-by-step workflows
• Added 2 agents with model/color/example frontmatter per plugin spec
• Added SessionStart hook for automatic project context injection
• Added README.md with plugin docs, structure, and testing strategy
• Updated root README.md with plugin installation, usage, and development docs
• Added Opus as a model option to the agent cdk stack source

Testing

46 checks executed, 45 passed, 1 failed (fixed in bab8794).

Testing strategy and how to reproduce: abca-plugin/README.md#testing

Acknowledgment

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of the project license.

[krokoko]

@rsmets are you still testing this ? Or we can switch it to ready for review ?

[rsmets]

@rsmets are you still testing this ? Or we can switch it to ready for review ?

I am still testing... I used it to get myself setup yesterday then had to switch gears. Now testing the post setup semantic interface.

[rsmets]

@krokoko thanks for keeping this branch's trunk up to date with main👍🏼

[rsmets]

Plugin Testing Results

Ran a 4-layer validation (structural, agent config, content integrity, hooks) across all plugin components.

46 checks executed, 45 passed, 1 failed (fixed in bab8794).

Testing strategy: abca-plugin/README.md#testing

Layer 1: Structural Validation — PASS (9/9)

Check	Result
plugin.json valid JSON	PASS
name field (kebab-case)	PASS
version field (semver)	PASS
description field	PASS
author field	PASS
Agent files discovered (2)	PASS
Skill files discovered (7)	PASS
Hook file discovered	PASS
No orphaned files	PASS

Layer 2: Agent Configuration — PASS (1 bug found and fixed)

cdk-expert.md — All PASS

agent-debugger.md — 1 issue:

• agent/prompts/ referenced but correct path is agent/src/prompts/ — fixed in bab8794

Layer 3: Content Integrity — PASS (25/25)

• 6/6 file paths exist (cli/lib/bin/bgagent.js, cdk/src/stacks/agent.ts, etc.)
• 10/10 mise tasks are valid (build, install, //cdk:compile, //cdk:test, etc.)
• 6/6 CLI commands verified against actual bgagent --help output
• 3/3 skill cross-references resolve to existing skills

Layer 4: Hook Validation — PASS (6/6)

Check	Result
Valid JSON structure	PASS
SessionStart event supported	PASS
Hook type prompt valid	PASS
All 7 skills listed exist	PASS
Both agents listed exist	PASS
No sensitive data	PASS

[krokoko]

Np thanks ! Just wanted to check in case you were waiting for me on an approval :)

[rsmets]

@krokoko I believe that this PR is now ready for review (and merge). Feel free to edit as you see fit.

[rsmets]

@krokoko I am actually running into 503 bedrock errors submitting a task in a fresh AWS account. I feel like this maybe model quota related (I was using Opus). I added some of this info the troubleshooting skill but unclear if that is really the root cause.

{
            "logStreamName": "trajectory/01KP75W2MKNBH82R5FYYDPQ6PY",
            "timestamp": 1776210863759,
            "message": "{\"event\": \"TRAJECTORY_TURN\", \"task_id\": \"01KP75W2MKNBH82R5FYYDPQ6PY\", \"turn\": 1, \"model\": \"<synthetic>\", \"thinking\": \"\", \"text\": \"API Error: 503 Too many connections, please wait before trying again.\", \"tool_calls\": [], \"tool_results\": []}",
            "ingestionTime": 1776210863766, "eventId": "39610825891551450069986979437098283030052058757434966016"
},

[rsmets]

Well good news... I got it up and running and interfacing with strictly through a semantic Claude code interface with the help of the plugin 👍🏼

i.e rsmets/personal-site#7