Learn

Codex compatibility note:

• Invoke repository skills with $skill-name in Codex; this mirrored copy rewrites legacy Claude /skill-name references.
• Task tracker mandate: BEFORE executing any workflow or skill step, create/update task tracking for all steps and keep it synchronized as progress changes.
• User-question prompts mean to ask the user directly in Codex.
• Ignore Claude-specific mode-switch instructions when they appear.
• Strict execution contract: when a user explicitly invokes a skill, execute that skill protocol as written.
• Subagent authorization: when a skill is user-invoked or AI-detected and its protocol requires subagents, that skill activation authorizes use of the required spawn_agent subagent(s) for that task.
• Do not skip, reorder, or merge protocol steps unless the user explicitly approves the deviation first.
• For workflow skills, execute each listed child-skill step explicitly and report step-by-step evidence.
• If a required step/tool cannot run in this environment, stop and ask the user before adapting.

Codex Project-Reference Loading (No Hooks)

Codex uses static project-reference loading instead of runtime-injected project docs. When coding, planning, debugging, testing, or reviewing, open project docs explicitly using this routing.

Always read:

• docs/project-config.json (project-specific paths, commands, modules, and workflow/test settings)
• docs/project-reference/docs-index-reference.md (routes to the full docs/project-reference/* catalog)
• docs/project-reference/lessons.md (always-on guardrails and anti-patterns)

Missing/stale context route: If docs/project-config.json, the docs index, lessons.md, CLAUDE.md, AGENTS.md, or any task-required reference doc is missing or stale, auto-run $project-init or the narrow setup route ($project-config, $docs-init, $scan-all, $scan --target=<key>, $claude-md-init) before ordinary project-specific work. If Codex mirrors or AGENTS.md are missing/stale, ask the user to run $sync-codex; do not auto-run it.

Situation-based docs:

• Project structure/architecture/tech-stack/deployment/setup (any layer — backend, frontend, or infra): project-structure-reference.md
• Backend/CQRS/API/domain/entity changes: backend-patterns-reference.md, domain-entities-reference.md
• Frontend/UI/styling/design-system: frontend-patterns-reference.md, scss-styling-guide.md, design-system/README.md
• Spec authoring, docs/specs/ pathing, or TC format: feature-spec-reference.md, spec-system-reference.md, spec-principles.md
• Behavior/public-contract changes or spec-test-code sync: workflow-spec-test-code-cycle-reference.md plus the spec docs above
• Derived spec indexes/ERDs/reimplementation guides: spec-system-reference.md and source Feature Specs under docs/specs/
• Integration test implementation/review: integration-test-reference.md
• E2E test implementation/review: e2e-test-reference.md
• Code review/audit work: code-review-rules.md plus domain docs above based on changed files

Do not read all docs blindly. Start from docs-index-reference.md, then open only relevant files for the task.

Quick Summary

Goal: Teach Claude lessons that persist across sessions by generalizing each lesson to its failure mode and saving it to the carrier a future session will actually read — the best-fit prose reference doc, or docs/project-config.json when the lesson is really a machine-readable project fact.

Summary: read-this-if-nothing-else digest of the main steps —

• Generalize before anything else — climb from the incident to the reusable failure mode; a lesson naming this ticket's files/services/tools is not a lesson yet.
• Triage (recurrence + auto-fix) BEFORE routing — a lesson a review skill already catches is noise.
• Classify the carrier, don't default to prose: a lesson stating a project FACT (path, run-command, module map, convention, tooling choice) belongs in docs/project-config.json, the machine-readable map every skill reads first; a lesson stating a RULE or pattern belongs in the matching docs/project-reference/ doc. Both can apply — write the fact to config AND the rule to prose. To learn what the config holds and its exact field names, read the file or use $project-config (it runs --describe).
• Assess prevention depth — doc/config update, prompt rule, static protocol lesson, hook, test, or skill update.
• Confirm target with the user, save, then run the 3 mandatory end tasks — Learn Review → $why-review → $prompt-enhance.

Workflow:

1. Capture -- Identify the lesson from user instruction or experience
2. Route -- Analyze lesson content against the Reference Doc Catalog AND docs/project-config.json, select the best target carrier (prose doc, config field, or both)
3. Save -- Append lesson to the selected file
4. Confirm -- Acknowledge what was saved and where
5. Learn Review -- Run the mandatory 2-step end gate (Learn Review + $why-review)
6. Enhance -- Run $prompt-enhance on modified file(s) to optimize AI attention anchoring

Key Rules:

• GENERALIZE FIRST (the #1 protocol): Extract the GENERIC lesson that applies to many cases — NEVER save the specific case as-is. The user's words describe one incident; your job is to climb from that incident to the reusable rule. Strip every project/file/tool/domain name. If the saved text only helps on this exact ticket, you failed — abstract it up a level. (Enforced by the Lesson Quality Gate below.)
• Triggers on "remember this", "always do X", "never do Y"
• Triage first: pass Recurrence gate + Auto-fix gate BEFORE routing or saving
• Smart-route to the most relevant file, NOT always docs/project-reference/lessons.md
• Consider docs/project-config.json on EVERY routing decision — a lesson that is really a project fact (path, run-command, module map, tooling choice) belongs in the machine-readable map, not in prose; read it or use $project-config to know its schema before deciding
• Use exact config schema field names (node .claude/hooks/lib/project-config-schema.cjs --describe) and prefer an existing field — NEVER invent a key, and route config writes through $project-config
• Check for existing entries before creating duplicates
• Confirm target file with user before writing

Be skeptical. Apply critical thinking, sequential thinking. Every claim needs traced proof, confidence percentages (Idea should be more than 80%).

Usage

Add a lesson

$learn always use the validation framework fluent API instead of throwing ValidationException
$learn never call external APIs in command handlers - use Entity Event Handlers
$learn prefer async/await over .then() chains

List lessons

$learn list

Remove a lesson

$learn remove 3

Clear all lessons

$learn clear

Reference Doc Catalog (READ before routing)

Each docs/project-reference/ file is auto-initialized by session-init-docs.cjs hook and populated by /scan-* skills. Understanding their roles is critical for correct routing: routing is static — read the doc whose Read Trigger matches your task.

Key insight: lessons.md and code-review-rules.md are the highest-recurrence routing targets — read them on every relevant task. Place high-recurrence lessons where the matching Read Trigger guarantees a future session opens them.

Also a routing destination: docs/project-config.json (machine-readable, NOT prose)

The catalog above is prose. docs/project-config.json is the project's machine-readable map — modules/paths, framework + search keywords, test/E2E/integration run-commands, design system, architecture rules, workflow patterns — and it is what CLAUDE.md and the docs/project-reference/** docs are generated from. Every skill is told to read it BEFORE investigating, planning, or coding, so a fact recorded there reaches more sessions than the same fact written into one prose doc.

Read Trigger: every task, ahead of the prose docs. Owned by: $project-config.

Learn its shape before routing anything into it — either read docs/project-config.json directly, or invoke $project-config, which knows the schema and its exact field names:

node .claude/hooks/lib/project-config-schema.cjs --describe   # exact field names + per-field derivation notes

Rules:

• MUST check docs/project-config.json as a candidate on EVERY routing decision — why: a project fact written only as prose is invisible to the tooling that reads the config, and is silently overwritten the next time the generated docs regenerate from it.
• MUST use exact schema field names (--describe, copy verbatim) and prefer an EXISTING field over a new one — why: unknown keys are accepted as warnings (project-config-schema.cjs:618,688), so an invented key looks like it worked while no consumer ever reads it.
• No existing field fits → do NOT invent a top-level key silently. Route the lesson to prose and surface the gap to the user as a proposed schema addition — why: a schema change is a framework decision, not a side effect of $learn.
• Prefer routing the config write through $project-config (Plan → Review → Execute + validation) over hand-editing the JSON — why: it validates after each phase and knows the field names this skill would otherwise guess.
• Config carries FACTS, never prose lessons — NEVER paste a narrative lesson into a config string field. — why: the config is consumed by tooling and by every skill's prefetch; prose there bloats every session and belongs in a reference doc.

Smart File Routing (CRITICAL)

Lesson Triage Gate (MANDATORY — run FIRST, before routing or saving)

Both gates must pass. A lesson review skills already catch adds noise without value. A one-off situational mistake won't be prevented by a persisted rule.

Routing Table

Route to the most relevant file based on lesson content:

Prevention Depth Assessment (MANDATORY before saving)

Before saving any lesson, critically evaluate whether a doc update alone is sufficient or a deeper prevention mechanism is needed:

Decision flow:

1. Capture the lesson
2. Ask: "Could this mistake recur if the AI forgets this lesson?" If yes → needs more than a doc update
3. Ask: "Can this be caught automatically by a test or hook?" If yes → recommend hook/test
4. Evaluate Static Protocol Lesson promotion (see below)
5. Present options to user with ask the user directly:
   • "Doc update only" — save to the best-fit reference file (default for most lessons)
   • "Doc + prompt rule" — also add to development-rules.md so all agents see it
   • "Doc + Static Protocol Lesson" — also add to shared protocol lessons (see criteria below)
   • "Full prevention" — plan a hook, test, or shared module to enforce it automatically
6. Execute the chosen option. For "Full prevention", create a plan via $plan instead of just saving.

Static Protocol Lesson Promotion (MANDATORY evaluation)

After generalizing a lesson, evaluate whether it qualifies as a Static Protocol Lesson in .claude/skills/shared/sync-inline-versions.md. Static protocol lessons are baked into CLAUDE.md, mirrored into AGENTS.md, and synced to Codex carriers through project-init/sync tooling.

Qualification criteria (ALL must be true):

1. Universal — Applies to ANY AI coding project, not just this codebase
2. High recurrence — AI agents make this mistake repeatedly across sessions without the reminder
3. Silent failure — The mistake produces no error/warning; it silently degrades output quality
4. Not already covered — No existing Static Protocol Lesson addresses the same root cause

Static Protocol Lessons — Universal AI mistake prevention rules baked into static carriers. Stored in .claude/skills/shared/sync-inline-versions.md under the ai-mistake-prevention and ai-mistake-prevention:full SYNC blocks. Each must be universal, high-recurrence, and silent-failure. READ .claude/skills/shared/sync-inline-versions.md to check for duplicates before adding.

If qualified: Recommend "Doc + Static Protocol Lesson" option. On user approval, append the lesson as a new bullet to the relevant shared SYNC blocks, then run the project-init / sync pipeline so CLAUDE.md, AGENTS.md, and Codex carriers regenerate from the shared source.

If NOT qualified: Explain why (e.g., "Too project-specific", "Already covered by existing Static Protocol Lesson about X", "Low recurrence — only happens in rare edge cases"). Proceed with doc-only or prompt-rule option.

Lesson Quality Gate (BLOCKING — generalize before you save)

CORE PROTOCOL — do not skip: A $learn request always arrives as a SPECIFIC case ("don't migrate via the bus and spam Elasticsearch"). Saving it verbatim is the default failure mode. You MUST transform specific → generic BEFORE writing: name the underlying class of mistake, drop the incident's nouns, and write a rule that fires across many future cases ("migrations write the DB directly, never via message bus — applies to all migrations"). If you cannot state the lesson without naming this ticket's files/services/tools, it is NOT generic yet — climb one more abstraction level. When in doubt, save the MORE generic version; a too-specific lesson is dead weight injected on every prompt.

Every lesson MUST be root-cause level and generic across any codebase. Apply this 3-step extraction before saving:

Step 1 — Name the FAILURE MODE, not the symptom:

The failure mode is the reasoning or assumption that broke — not what the output looked like.

Step 2 — Verify generality:

Does this failure mode apply to ≥3 different contexts or codebases? If only one file or one specific case → go up one abstraction level. A good lesson prevents an entire class of mistakes.

Step 3 — Write as a universal rule:

• Strip ALL project-specific names, file paths, class names, and tool names
• Must be useful on any codebase, any language, any task type
• If multiple mistakes share the same failure mode → consolidate to ONE lesson, not many
• Test: "Would an AI working in Java, Go, or Python on a different project benefit from this?" If yes → good. If no → rewrite.

Anti-pattern examples:

• BAD: "Always check lib/dedup-constants.cjs for marker strings" → project-specific path
• GOOD: "When consolidating modules, ensure shared constants are imported from a single source of truth — never define inline duplicates."
• BAD: "Update .claude/docs/hooks/README.md after deleting hooks" → project-specific file
• GOOD: "Deleting components causes documentation staleness cascades — map all referencing docs before removal."
• BAD: "Read GlobalUsings.cs before adding usings in *.IntegrationTests" → project-specific file
• GOOD: "Before generating code that uses project conventions (imports, namespaces, annotations), read the project's bootstrap/configuration files for that layer — convention files override framework defaults silently."

End-Phase Learn Review Gate (MANDATORY before marking complete)

Run these 2 tasks at the end of every $learn operation:

Task 1 — Learn Review (value + generality + recurrence):

• Keep only lessons with clear prevention value.
• Lesson must be either:
  • Universal across many projects/codebases, OR
  • A stable project-wide principle (architecture invariant, naming invariant, workflow invariant).
• Reject lessons that are:
  • Specific to the current ticket/change/file,
  • Rare edge cases with low recurrence,
  • Already covered by existing lessons or review skills.
• If target is docs/project-reference/lessons.md (injected on every prompt), apply stricter bar: high impact + high recurrence only.

Task 2 — Run $why-review (adversarial challenge):

• Use $why-review to challenge whether this lesson deserves persistent memory.
• Verify:
  • Why this lesson prevents repeated mistakes,
  • Why this should be a lesson instead of a one-time note,
  • Why auto-checks ($code-review, $code-simplifier, $security-review, $lint, hook/test) are insufficient.
• If rationale is weak, rewrite at higher abstraction or skip $learn.

Routing Decision Process

1. Run Triage Gate — recurrence + auto-fix filters; stop here if either fails
2. Read the lesson text — identify keywords and domain
3. Apply Lesson Quality Gate — analyze root cause, generalize, verify universality
4. Classify the carrier — FACT vs RULE (do this BEFORE the Routing Table). Ask: "Is this a machine-readable project fact, or a rule an agent must reason with?" Fact → docs/project-config.json; rule → a prose reference doc; both → both. To decide, read the config or use $project-config (--describe) so the judgment rests on the real schema, never on a guess about what the config holds. — why: skipping this step is how a project fact ends up as prose that no tooling reads and the next regeneration contradicts.
5. Run Prevention Depth Assessment — determine if doc/config-only or deeper prevention needed
6. Match against Routing Table — pick the best-fit file (or config field)
7. Tell the user: "This lesson fits best in docs/{file}. Confirm? [Y/n]"
8. On confirm — read target file, find the right section, append the lesson (config target → route through $project-config)
9. On reject — ask user which file to use instead

Format by Target File

For docs/project-reference/lessons.md (general lessons):

- [YYYY-MM-DD] <lesson text>

For pattern/rules files (code-review-rules, backend-patterns, frontend-patterns, integration-test):

• Find the most relevant existing section in the file
• Append the lesson as a rule, anti-pattern entry, or code example
• Use the file's existing format (tables, code blocks, bullet lists)
• If no section fits, append to the Anti-Patterns or general rules section

Budget Enforcement (MANDATORY for docs/project-reference/lessons.md)

docs/project-reference/lessons.md is a static project-reference carrier read during project work. Token budget must be controlled.

Hard limit: 20000 characters (~6666 tokens). Check BEFORE saving any new lesson.

Workflow when adding to docs/project-reference/lessons.md:

1. Read file, count characters (wc -c docs/project-reference/lessons.md)
2. If current + new lesson > 20000 chars → trigger Budget Trim before saving
3. If under budget → save normally

Budget Trim process:

1. Display all current lessons with char count each
2. Evaluate each lesson on two axes:
   • Universality — How often does this apply? (every session vs rare edge case)
   • Recurrence risk — How likely is the AI to repeat this mistake without the reminder?
3. Score each: HIGH (keep as-is), MEDIUM (candidate to condense), LOW (candidate to remove)
4. Present to user with ask the user directly: "Budget exceeded. Recommend removing/condensing these LOW/MEDIUM items: [list]. Approve?"
5. On approval: condense MEDIUM items (shorten wording), remove LOW items, then save new lesson
6. On rejection: ask user which to remove/condense

Condensing rules:

• Remove examples, keep the rule: "Patterns like X break Y syntax" → just state the rule
• Merge related lessons into one if they share the same root cause
• Target: each lesson ≤ 250 chars (one concise sentence + bold title)

Does NOT apply to: Other routing targets (backend-patterns-reference.md, code-review-rules.md, etc.) — those files have their own size and are injected contextually, not on every prompt.

Behavior

1. $learn <text> — Route and append lesson to the best-fit file (check budget if target is lessons.md)
2. $learn list — Read and display lessons from ALL 12 target files (show file grouping + char count for lessons.md)
3. $learn remove <N> — Remove lesson from docs/project-reference/lessons.md by line number
4. $learn clear — Clear all lessons from docs/project-reference/lessons.md only (confirm first)
5. $learn trim — Manually trigger Budget Trim on docs/project-reference/lessons.md
6. File creation — If target file doesn't exist, create with header only

Auto-Inferred Activation

When Claude detects correction phrases in conversation (e.g., "always use X", "remember this", "never do Y", "from now on"), this skill auto-activates. When auto-inferred (not explicit $learn), confirm with the user before saving: "Save this as a lesson? [Y/n]"

How Lessons Reach the AI

Lessons and pattern references are read statically, per the project-reference-docs gate in CLAUDE.md:

• docs/project-reference/lessons.md — read on every task (the gate always includes it).
• Pattern/rule references (backend-patterns-reference.md, code-review-rules.md, etc.) — read by their matching trigger (see the Reference Doc Catalog table above).

Because the routing is static prose, hookless harnesses (Codex) load the same lessons and patterns as Claude Code.

Prompt Enhancement (MANDATORY final step)

After saving a lesson to any target file, run $prompt-enhance on the modified file(s) to optimize AI attention anchoring and token quality.

When to run:

• After EVERY successful lesson save (regardless of target file)
• Pass the specific file path(s) that were modified

What it does:

• Ensures the new lesson integrates with existing top/bottom summary anchoring
• Optimizes token usage — tightens prose, merges redundant content
• Verifies no content loss from the save operation

How to invoke:

$prompt-enhance docs/project-reference/<modified-file>.md

Skip conditions (do NOT run prompt-enhance if):

• The save was to lessons.md AND the file is under 1500 chars (too small to benefit)
• The user explicitly requests "save only, no enhance"

[IMPORTANT] Use task tracking to break ALL work into small tasks BEFORE starting — including tasks for each file read. This prevents context loss from long files. For simple tasks, AI MUST ATTENTION ask user whether to skip.

Mandatory end tasks are ALWAYS (in order):

1. "Run Learn Review (lesson value + generality + recurrence gate)."
2. "Run $why-review to challenge whether the lesson is worth persistent memory."
3. "Run $prompt-enhance <modified-file> to optimize lesson content for AI attention anchoring."

Do NOT mark the skill complete until all 3 tasks run.

AI Mistake Prevention — Failure modes to avoid on every task:

Re-read files after context changes. Context compaction, resume, or long-running work can make memory stale; verify current files before acting. Verify generated content against source evidence. AI hallucinates APIs, names, claims, and document facts. Check the relevant source before documenting or referencing. Check downstream references before deleting or renaming. Removing an artifact can stale docs, generated mirrors, configs, and callers; map references first. Trace the full impact chain after edits. Changing a definition can miss derived outputs and consumers. Follow the affected chain before declaring done. Verify ALL affected outputs, not just the first. One green check is not all green checks; validate every output surface the change can affect. Assume existing values are intentional — ask WHY before changing OR flagging one as a defect. Before changing or reporting a constant, limit, flag, cutoff, wording, or pattern, read nearby context and history, the CALLER's ordering, and 2+ sibling call sites of the same convention. A doc stating WHAT without WHY is missing rationale, not proof of a missing guard. Surface ambiguity before acting — don't pick silently. Multiple valid interpretations require an explicit question or stated assumption with risk. Assert the outcome your system owns, not the intermediate state your infrastructure owns. When verifying async work, assert the final business state — never the delivery/retry bookkeeping held in shared infrastructure that any co-running process can write. Such a check passes when run alone and flakes the moment anything else shares that infrastructure. Keep shared guidance role-relevant. Universal guidance must help every receiving skill or agent; code-specific obligations belong only in code-specific protocols.

Critical Thinking Mindset — Apply critical thinking, sequential thinking. Every claim needs traced proof, confidence >80% to act. Anti-hallucination: Never present guess as fact — cite sources for every claim, admit uncertainty freely, self-check output for errors, cross-reference independently, stay skeptical of own confidence — certainty without evidence roo

…(truncated)