Cc Best Practice Adoption

CC Best Practice Adoption

Workflow for analyzing external Codex best-practice sources, distilling actionable gaps in our ecosystem, and wiring daily learning into the session lifecycle.

When to Use

• New Codex version ships with features to evaluate
• Reviewing an external repo (e.g., shanraisshan/Codex-best-practice)
• Expanding the daily learning tip catalog
• Bootstrapping native .Codex/ structure (agents, commands, skills)

Gotchas

1. Native vs Custom — Check Before Building

Critical lesson learned: /powerup, /insights, /stats, /context, /simplify, /batch, /debug, /diff, /security-review are all native CC commands (v2.1.90+). We wasted effort planning to build custom versions before discovering they exist.

Rule: Before creating any slash command, run /help or check Codex-best-practice/best-practice/Codex-commands.md for the 64 native commands.

2. Hermes Skills vs Native CC Skills

Our ecosystem has 691 Hermes-style skills but ZERO native CC format. They are different:

• Native CC skills: .Codex/skills/<name>/SKILL.md with CC frontmatter (context: fork, paths, model, effort, hooks)
• Hermes skills: ~/.hermes/skills/<name>/SKILL.md with Hermes frontmatter
• Both can coexist. Native CC skills show in /skills menu and support auto-discovery.

3. Tips Catalog YAML Is Not Standard YAML

config/workflow-tips/tips-catalog.yaml uses a minimal structure parsed by daily-learning.py without PyYAML. Don't add complex YAML features (anchors, flow sequences in values). Keep entries as simple key: value pairs.

4. Deterministic Daily Seed

daily-learning.py uses md5(YYYY-MM-DD) as seed so the same 2 tips show all day. Running it multiple times won't cycle tips — that's intentional (spaced repetition).

Workflow Steps

Step 1: Clone and Analyze External Source

cd /tmp && git clone --depth 1 <repo-url>

Key files to read first:

• README.md (feature catalog, tip index)
• best-practice/Codex-commands.md (native commands list)
• best-practice/Codex-skills.md (native skills + frontmatter fields)
• best-practice/Codex-subagents.md (native agent frontmatter fields)
• best-practice/Codex-settings.md (60+ settings, 170+ env vars)

Step 2: Gap Analysis

Compare external catalog against:

1. config/workflow-tips/tips-catalog.yaml — what tips are we missing?
2. .Codex/settings.json — what settings are we not using?
3. .Codex/agents/ — do we have native agent definitions?
4. .Codex/commands/ — do we have native command definitions?

Output a structured list: ALREADY DONE / STILL MISSING / NEW TO ADD.

Step 3: Expand Tips Catalog

Append new tips to config/workflow-tips/tips-catalog.yaml following the format:

  - id: <prefix>-<slug>      # cc- | eco- | gsd- | bp-
    category: <category>      # Codex | ecosystem | gsd | practice
    name: <Short Name>
    oneliner: "<one sentence>"
    try_it: "<copy-pasteable command>"
    source: "<provenance>"
    tags: [tag1, tag2]
    added: YYYY-MM-DD

Prefixes: cc- (Codex native), eco- (ecosystem), gsd- (GSD), bp- (best practice).

Step 4: Update Daily Learning Script

If new tip categories are added, update scripts/productivity/daily-learning.py:

• Add category label mapping in show_daily_tips()
• Add practice exercises in generate_practice() for high-value tips

Step 5: Bootstrap Native .Codex/ Structure

Create files using native CC frontmatter (not Hermes format):

Agents (.Codex/agents/<name>.md):

---
name: <name>
description: "<when to invoke — use PROACTIVELY for auto-invocation>"
model: haiku|sonnet|opus
tools: Read, Write, Edit, Bash, Glob, Grep
color: cyan|red|green|blue|magenta
memory: project
effort: low|medium|high|max
isolation: worktree    # optional
skills:                # optional — preloaded into context
  - skill-name
---

Commands (.Codex/commands/<name>.md):

---
name: <name>
description: "<what it does>"
model: haiku|sonnet|opus
effort: low|medium|high
allowed-tools: Read, Bash, Agent(explorer)
---
Use !`command` for dynamic shell injection — output injected into prompt.

Settings (.Codex/settings.json) — key fields often missed:

• outputStyle: "Explanatory" — insight boxes
• effortLevel: "high" — default reasoning depth
• worktree.symlinkDirectories — fast worktree startup
• spinnerVerbs — domain-relevant waiting messages
• spinnerTipsOverride — ecosystem tips while spinning

Step 6: Wire Into Session Lifecycle

Add a SessionStart hook for automatic tip surfacing:

{
  "type": "command",
  "command": "bash .Codex/hooks/daily-learning-tip.sh 2>/dev/null || true",
  "timeout": 3,
  "statusMessage": "Daily learning tip"
}

Step 7: Track Progress

uv run scripts/productivity/daily-learning.py --progress    # category coverage
uv run scripts/productivity/daily-learning.py --categories  # browse all tips

Key Files

Reference

• https://github.com/shanraisshan/Codex-best-practice (133k+ stars)
• 64 native CC commands, 5 official skills, 5 official agents
• CC frontmatter docs: commands, skills, subagents pages in best-practice repo
• GitHub issues: #1775 (8-week cadence), #1760 (self-improvement commands)
• https://github.com/affaan-m/everything-Codex (50K stars, MIT)
  • Cherry-pick: security guide, pre:config-protection hook, batch-at-Stop pattern, context-budget skill, autonomous-loops (6 patterns), continuous-learning-v2 instincts
  • Implemented pattern: add a dedicated Codex PreToolUse hook entry in .Codex/settings.json for Write|Edit|MultiEdit, then delegate the decision to a reusable shell checker (example: .Codex/hooks/config-protection-pretooluse.sh -> scripts/enforcement/check-config-protection.sh). This keeps policy logic testable outside the hook wrapper.
  • Practical allowlist learned in implementation: permit non-tooling metadata edits in pyproject.toml and other low-risk config touches, but block broad weakening patterns like ignore, extend-ignore, per-file-ignores, disable, skip, off, and block removal of core safety-gate text from AGENTS.md / AGENTS.md unless an explicit env bypass is set.
  • Testing pattern: write focused pytest coverage for the hook wrapper behavior before implementation, including non-protected file pass-through, allowlisted metadata edits, blocked risky config edits, blocked safety-gate removal, and explicit bypass env (CONFIG_PROTECTION_APPROVED=1).
  • MCP servers to evaluate: evalview, insaits, token-optimizer, omega-memory
  • ~80% overlap with our 691 skills — adopt patterns not bulk content

Version History

• 1.0.0 (2026-04-03): Initial — gap analysis workflow, tips expansion, native CC structure bootstrap