lcs-chain-of-truth

LCS Chain of Truth

Purpose

lcs-chain-of-truth is a meta-skill — it does not replace other skills. It is a protocol injected by other LCS skills when they need to produce auditable, verifiable, traceable evidence of their work.

Use it as a layer added on top of skill execution, not as a standalone flow.

---

Trigger

Activate this protocol when any LCS skill:

• Produces output that will be committed or shipped
• Makes decisions that affect contracts, specs, or plans
• Executes code changes, reviews, or finalizations
• Requires a verifiable audit trail for human review

---

Chain of Truth Flow

Every Chain of Truth report must traverse these 6 stages in order:

Source → Assumption → Plan → Action → Verification → Report

Stage	Description
Source	Identify the raw inputs: files read, user inputs, prior artifacts, external references. Quote exact paths and line numbers.
Assumption	List every assumption made about inputs, context, or behavior. Flag assumptions that could not be verified.
Plan	State the intended actions before executing them. No hidden reasoning.
Action	Record what was actually done — commands run, files written, edits applied.
Verification	Run available checks. Record exact command output. Use fallbacks if verification tools unavailable.
Report	Summarize: what passed, what failed, what remains unverified, and confidence level.

---

Chain of Thought Warning

Prohibition: Do NOT expose hidden internal chain-of-thought in outputs.

Chain of Truth is about external, auditable evidence — sources, actions, and verifiable results. It is not a dump of internal reasoning steps.

• Allowed: "Read file X line 12. Found pattern Y. Wrote Z."
• Prohibited: "I think this probably works because intuitively…"
• Prohibited: Claiming verification passed without running it.
• Prohibited: Skipping source identification.

---

Four Levels

Choose the level based on risk and context. Declare it at the start of the Chain of Truth Report.

Level 1 — Light

Use for: Exploratory or conversational skills (lcs-explore, early brainstorming).

Requirements:

• Source: identify key inputs
• Assumption: list top 2–3 assumptions
• Plan: brief intent statement
• Action: summary of what was done
• Verification: manual check or "not applicable"
• Report: 1–3 sentences

Level 2 — Standard

Use for: Planning and documentation skills (lcs-new, lcs-toprd, lcs-onboarding, lcs-task-slicer, lcs-self-improvement).

Requirements:

• Source: all files read, with paths
• Assumption: full list, flagging unverified ones
• Plan: enumerated steps
• Action: per-step record of what was done
• Verification: command output or manual review result
• Report: structured summary with confidence rating

Level 3 — Strict

Use for: Contracts, reviews, finalization (lcs-prd-reviewer, lcs-doc-finalizer, lcs-codebase-doc).

Requirements:

• All Level 2 requirements, plus:
• Source: exact line references for all quoted content
• Assumption: each assumption must be labeled [verified] or [unverified]
• Verification: must attempt automated checks; document every failure
• Report: explicit pass/fail per acceptance criterion

Level 4 — Very Strict

Use for: Code changes, debugging, implementation (lcs-task-executor, lcs-debug-ext).

Requirements:

• All Level 3 requirements, plus:
• Action: exact commands run with stdout/stderr captured
• Verification: all test and lint commands must be run; results quoted verbatim
• Report: blocked items listed explicitly; no "assumed passing" allowed
• Confidence must be stated as high, medium, or low with justification

---

Report Placement Rule

The Chain of Truth Report section must appear before the ## Handoff section in any artifact that uses this protocol.

## Chain of Truth Report
Level: <1–4>
...

## Handoff
...

---

Verification Rules

1. Attempt verification if tools are available: run linter, test suite, or Test-Path as appropriate.
2. Fallbacks for markdown/docs repos: if no test suite exists, use Select-String pattern checks or manual read confirmation.
3. Never skip: record "not applicable" with a reason rather than silently omitting verification.
4. Never claim pass without running: if a check was not executed, mark it [unverified].

---

Anti-Patterns

Anti-Pattern	Why Prohibited
Claiming verification passed without running it	Breaks auditability
Exposing internal chain-of-thought reasoning	Pollutes the evidence chain with speculation
Skipping source identification	Makes the report unreproducible
Writing Report before Action completes	Report must reflect reality, not intent
Omitting failed checks from Report	Hides risk from reviewers

---

Skill Level Mapping

Each LCS skill declares its Chain of Truth level:

Skill	Level	Rationale
lcs-explore	Light (1)	Exploratory, conversational, low-stakes
lcs-toprd	Standard (2)	Planning artifact, structured output
lcs-tosrs	Strict (3)	Spec derivation from PRD
lcs-onboarding	Standard (2)	Documentation generation
lcs-task-slicer	Strict (3)	Task breakdown planning
lcs-self-improvement	Standard (2)	Diagnostic analysis and recommendations
lcs-prd-reviewer	Strict (3)	Reviews contracts, flags acceptance criteria
lcs-doc-finalizer	Strict (3)	Finalizes and archives canonical docs
lcs-codebase-doc	Strict (3)	Codebase mapping and onboarding documentation
lcs-debug	Standard (2)	Question-asking investigation skill, no code changes applied
lcs-debug-ext	Very Strict (4)	Evidence-based debug reports without code changes
lcs-task-executor	Very Strict (4)	Code implementation with tests and lint (canonical spelling)
lcs-new	Standard (2)	Blank work-item registration, no artifacts created
lcs-chain-of-truth	—	Meta-protocol, not self-applied

---

Chain of Truth Level

Level: Meta

This skill defines the LCS Chain of Truth protocol and is not self-applied.

Handoff

Next recommended skill: lcs-shared Next file to read: skills/lcs-shared/contract.md Current phase: reference Current confidence: high Blocking questions: None Risks to carry forward: None Source of Truth Bundle: skills/lcs-chain-of-truth/SKILL.md (reference, not self-applied) Must Preserve IDs: None Unresolved IDs: None Suggested next command: N/A — reference skill, no execution flow