CategoryWeightCategory scoreContribution

Securityprompt, exec, net, exfil, eval

35%

7.0 pts

Supply chainhash, typosquat, maintainer, lockfile

20%

100

20.0 pts

Maintenancestaleness, pinning, CI

15%

100

15.0 pts

TransparencySKILL.md, perms, README

15%

100

15.0 pts

Communityinstalls, verify, response

15%

100

15.0 pts

Findings & checks · 1 flagged

Securityscore 20 · 1 finding

CRITICALReads your AWS credentials fileSS-PLUGIN-SECRET-EXFIL-AWS-FILES-01 · Credential exfiltration · scripts/secret-scan.js

CRITICAL — full cloud credentials are tier-1 exfiltration material — a read here can mean total account takeover.

Why it matters

This plugin references the AWS credentials file or the access-key fields stored inside it ({ name: 'AWS Secret Access Key', re: /\baws_secr…). Those are long-lived keys with broad cloud access, so any code that reads them can hand your whole AWS account to whatever it contacts next.

The exact value spotted

excerptscripts/secret-scan.js· javascript

10const CONFIDENT = [

11{ name: 'AWS Access Key ID', re: /\bAKIA[0-9A-Z]{16}\b/ },

12{ name: 'AWS Secret Access Key', re: /\baws_secret_access_key\s*=\s*['"][A-Za-z0-9/+=]{40}['

… (7 chars elided on L12)

13{ name: 'GitHub Token (classic)', re: /\bghp_[A-Za-z0-9]{36,}\b/ },

14{ name: 'GitHub Token (fine-grained)', re: /\bgithub_pat_[A-Za-z0-9_]{80,}\b/ },

Occurrences

1 occurrence · at L12

How to fix

Remove the direct read of ~/.aws/credentials; let the AWS SDK resolve credentials through its standard provider chain instead.

Delete code that opens or parses the credentials file or its key fields by hand.
Use the SDK's default credential resolution so secrets never pass through plugin code or leave the machine.

Avoid

creds = open(os.path.expanduser("~/.aws/credentials")).read()
requests.post(url, data={"creds": creds})

Safer pattern

# let the SDK resolve credentials; never read or transmit the file yourself
import boto3
s3 = boto3.client("s3")

Framework references

OWASPLLM02 ATLASAML.T0025 CWE-200

Trace & refs

ruleSS-PLUGIN-SECRET-EXFIL-AWS-FILES-01sha256be2351089858bb9f…rubric 365aacaView on GitHub

Supply chainscore 100 · 0 findings

✓—All supply chain checks passedNo findings in this category for the latest scan.pass

Maintenancescore 100 · 0 findings

✓—All maintenance checks passedNo findings in this category for the latest scan.pass

Transparencyscore 100 · 0 findings

✓—All transparency checks passedNo findings in this category for the latest scan.pass

Communityscore 100 · 0 findings

✓—All community checks passedNo findings in this category for the latest scan.pass

The primary manifest — the file an agent reads to learn what this artifact does.

README.md · 1 file

README.md10.2 KB · Markdown

Holocron

A Claude Code plugin that gives you a Jedi council of role-specialist agents, domain commands (a11y, perf, QA, security, ADR), and a SQLite-backed learning crystal that compounds your corrections across sessions.

A holocron, in Star Wars lore, is a crystalline cube that stores Jedi teachings. This one stores yours — every correction you make becomes durable knowledge the assistant carries into the next session.

[atuljha23.github.io/holocron](https://atuljha23.github.io/holocron/) · v0.3.0 · MIT · targets SWEs, frontend engineers, backend engineers, QAs, SREs, and security.

Why holocron

If you use Claude Code daily you already know the failure mode: you correct the assistant the same way a dozen times, every session feels like starting over, and the "plan → review" discipline rots the moment compaction hits. Holocron does three things about that:

Specialist agents — not one generic "reviewer", but ten role-flavored agents (@frontend-engineer, @qa-engineer, @a11y-auditor, @security-reviewer, @perf-engineer, …) each with a sharpened rubric and clear handoff boundaries.
Domain commands — a11y audits, performance checks, test-gap analysis, flaky-test hunting, security scans, ADR drafting. The stuff engineers actually reach for, as slash commands.
The crystal — a SQLite + FTS5 store of your corrections. New sessions load the top matches automatically. Compaction can't wipe them.

Install (30 seconds)

# Add this repo as a marketplace and install the plugin
/plugin marketplace add atuljha23/holocron
/plugin install holocron@holocron

# One-time: install the native dependency the learning crystal needs
cd ~/.claude/plugins/marketplaces/holocron && npm install

Local dev against a clone:

git clone https://github.com/atuljha23/holocron ~/projects/holocron
cd ~/projects/holocron && npm install
/plugin marketplace add ~/projects/holocron
/plugin install holocron@holocron

Health check:

/holocron:doctor

What's inside (v0.3.0)

Component	Count	Where
Role / meta agents	14	`agents/`
Slash commands	24	`commands/`
Skills	11	`skills/`
Hook scripts	13	`scripts/`
Hook events wired	6	`hooks/hooks.json`
Context profiles	4	`contexts/`
Rules	4	`rules/`

Run node scripts/holocron-cli.js inventory anytime to verify counts match this table.

v0.3.0 — Token Economy. Every Claude Code session's $ and tokens are logged to ~/.holocron/sessions/. Run /holocron:cost for the rolling total, /holocron:budget to set caps, /holocron:mcp-audit to cut schema bloat, @cost-analyst for ranked savings proposals.

Quickstart by role

If you're a software engineer

/holocron:onboard            # draft a CLAUDE.md for this repo
/holocron:plan               # enter plan mode with a real scaffold
/holocron:develop <feature>  # research → plan → implement → review
/holocron:commit             # quality-gated commit with a reviewed diff
/holocron:learn "<rule>"     # capture a correction so it sticks

If you're a frontend engineer

@frontend-engineer <task>    # role-specialist for component/UX work
/holocron:a11y               # audit changed UI files against WCAG 2.2 AA
/holocron:perf               # bundle / LCP / CLS check on the diff
/holocron:test-gap <file>    # find the component states nobody tests

If you're a backend engineer

@backend-engineer <task>     # API / data / service work
/holocron:sec-scan           # SAST + secrets + dep advisories on the diff
/holocron:perf               # N+1 and index-gap sweep
/holocron:adr "<title>"      # record a real decision, not a Slack message

If you're in QA

@qa-engineer <task>          # test strategy / coverage / flakes
/holocron:test-gap <file>    # prioritized list of missing tests
/holocron:flaky <test>       # walk the flake-triage tree with evidence

If you're SRE / platform

@sre-engineer <task>         # Docker / k8s / CI / runbooks
/holocron:llm-gate "push to main"
/holocron:permission-tune    # reduce prompt fatigue with allow/deny tuning

If you're security

@threat-modeler <feature>    # STRIDE on a proposed change
/holocron:threat-model       # design-level review
/holocron:sec-scan           # SAST + secrets + CVE on the diff

If you care about cost

/holocron:cost --breakdown          # $ + tokens + cache hit rate
/holocron:budget --session 2 --daily 10
/holocron:mcp-audit                 # MCP token overhead
@cost-analyst                       # ranked savings from usage data

Agents

All live under agents/ and invoke as @<name>:

Agent	Use when
`@architect`	Picking between approaches, decomposing a feature, writing an ADR
`@frontend-engineer`	React/Vue/Svelte/Web Components, state, forms, rendering
`@backend-engineer`	APIs, services, data layer, async patterns
`@qa-engineer`	Test strategy, coverage gaps, flakes, BDD
`@a11y-auditor`	WCAG 2.2 AA audits on UI components
`@perf-engineer`	Core Web Vitals, bundle size, N+1, index gaps
`@security-reviewer`	OWASP review, authn/authz, injection surfaces
`@threat-modeler`	STRIDE / LINDDUN design-level threat modeling
`@sre-engineer`	Dockerfile / k8s / CI / SLOs / runbook drafting
`@data-engineer`	SQL review, dbt models, pipelines, schema drift
`@cost-analyst`	Ranked $-saving proposals from session usage data
`@debugger`	Hypothesis-driven root-cause analysis
`@code-reviewer`	Staff-level PR review
`@docs-writer`	READMEs, runbooks, ADRs, API docs

Each agent's system prompt includes a "when NOT to use me — delegate to X" section so they don't step on each other's toes.

Commands

Namespaced as /holocron:<name>:

Meta / workflow: onboard, plan, develop, commit, review, handoff, learn, recall, insights, worktree, doctor

Domain: a11y, perf, test-gap, flaky, sec-scan, threat-model, adr

Ops / gates: permission-tune, llm-gate

Token economy: cost, budget, mcp-audit, context-size

Each command file documents its argument hints and delegates to the right agent(s).

Skills

Cross-cutting knowledge, auto-loaded when relevant:

context-engineering — Write / Select / Compress / Isolate
test-patterns — deep assertions > spies, AAA, table-driven
api-design — REST + GraphQL conventions, pagination, idempotency
a11y-checklist — WCAG 2.2 AA walkthrough
security-checklist — OWASP-style pre-merge sweep
perf-checklist — Core Web Vitals + backend query rubric
debug-playbook — hypothesis-driven triage
pr-review-rubric — what a staff engineer actually looks at
observability-checklist — logs / traces / metrics / correlation
migration-patterns — expand / contract, backfill, online DDL
token-economy — model tiering, cache hygiene, Read discipline, MCP pruning

Hooks

Six events, thirteen scripts. Everything fails safe — hooks never crash the session.

Event	What it does
`SessionStart`	Injects top-K recent learnings from the crystal as additionalContext
`UserPromptSubmit`	FTS-matches the prompt against learnings, injects top 3
`PreToolUse` Edit\	Write	Read-before-write tracker + regex secret scanner (blocks on confident matches)
`PreToolUse` Read	Nudges `offset` + `limit` on files > 500 lines
`PreToolUse` Bash(git commit*)	Surfaces repo-detected quality gates
`PreToolUse` Bash(git push*)	Nudges `/holocron:handoff` if none recorded
`PostToolUse` Edit	Sniffs `console.log`/`debugger`/`print`/`TODO` debug residue
`PostToolUse` Bash(test)	On test failure, nudges `/holocron:learn`
`PostToolUse *`	Flags tool responses > ~8k tokens (context bloat)
`PreCompact`	Snapshots session state, advises on post-compact recall
`Stop`	Parses the transcript and writes a per-session $ / token record; checks budget; nudges correction capture

The learning crystal

A SQLite database with an FTS5 index. Two scopes:

user — ~/.holocron/learnings.db, follows you across all projects
project — <repo>/.holocron/learnings.db, lives with the repo (commit or ignore per team)

Add, search, list:

/holocron:learn "no db mocks" --reason "mock/prod drift burned us on the Q1 migration" --tags testing,db
/holocron:recall flaky
/holocron:recall database

Or directly:

node scripts/holocron-cli.js add --title "no db mocks" --rule "integration tests hit a real db" --tags testing,db
node scripts/holocron-cli.js list --limit 20
node scripts/holocron-cli.js search "flaky"

Schema and triggers live in scripts/holocron-db.js — one file, zero build step, easy to fork.

Configuration examples

mcp-config.example.json — curated MCP servers (context7, playwright, github, postgres)
settings.example.json — permissions allowlist/denylist tuned for holocron + sensible auto-compaction

Neither is loaded automatically — copy what you want into your own Claude Code settings.

Requirements

Claude Code (latest)
Node ≥ 18 (the hook scripts and learning crystal are Node-based)
better-sqlite3 — installed via npm install in the plugin dir

Acknowledgements

Structural inspiration: rohitg00/pro-workflow (MIT). Holocron reimplements the persistent-learning idea from scratch (no code copied) and builds role-specialist content on top. If you want the workflow-meta focus, go check out pro-workflow — different tool for a different shape of user.

Plugin authoring reference: Anthropic's Claude Code plugin docs.

License

MIT — see LICENSE.

Contributing

v0.1.0 is intentionally scoped. Planned for v0.2.0:

Visual-regression / Storybook integration
Mobile agents (React Native, Swift, Kotlin)
Data-engineering agents (pipelines, SQL)
Terraform / Helm / Dockerfile reviewers
Load testing and trace inspection commands

Open an issue or PR.

Holocron— plugin

Is Holocron safe to install?

Holocron

Why holocron

Install (30 seconds)

What's inside (v0.3.0)

Quickstart by role

If you're a software engineer

If you're a frontend engineer

If you're a backend engineer

If you're in QA

If you're SRE / platform

If you're security

If you care about cost

Agents

Commands

Skills

Hooks

The learning crystal

Configuration examples

Requirements

Acknowledgements

License

Contributing

Audit the pieces. Scan the whole. Decide.