v0.1.0

v0.1.0 — Core Principles

Released: 2026-01 Theme: Name the discipline. Count failures. Index the project.

TL;DR

The first release. Establishes the eleven core principles the framework still runs on, introduces the mission counter with Wave 1 / Wave 2 escalation, and defines the สารบัญ (index) system: .kasidit/INDEX.md + RELATIONS.md + MEMORY.md.

Why

Observed pattern: senior engineers reason from mental models of the real system; AI pattern-matches surface similarity. The fix is not a bigger model — it is discipline externalized. Name the discipline, make it operational, count the failures so spirals end.

What's new

Eleven core principles

1. Mission-driven, not exploration-driven — every session has one mission; vague asks get narrowed.

2. Minimal requirement — narrowest interpretation; don't add features unasked.

3. Design before code — for anything multi-file, show endpoint / data flow / schema / relation diagram first; wait for user confirm.

4. Official documentation only (expanded in v0.2.1 into the Trust Hierarchy) — memory is not truth.

5. Check environment before generating code — language version, framework version, binary path.

6. Step one by one — state, apply, verify, report, next.

7. Focus only the current step — evict context for unrelated steps; drop failed attempts.

8. Explain = Hallucinate — reply in points, not paragraphs; no preamble; no recap.

9. Runtime is the judge — code compiles / lints / has tests → not done. Code runs with expected output → done.

10. Problem is in data, not in UI — 90% of "UI bugs" are data bugs. Check payload before editing UI layer.

11. Confidence labels (mandatory form from v0.3.0) — [high / medium / low / unsure].

Mission counter

Each mission has a failure counter. Failure = runtime did not pass.

• Count 1–4: AI retries, each with a different hypothesis
• Count 4 → Wave 1: stop generating fixes; ask user for bullet points on expected outcome
• Count 8 → Wave 2: stop entirely; output current state and all attempted fixes; ask user to redefine

What counts as failure: runtime error, expected vs actual mismatch, "still not working".

What does not count: user typed command wrong, cache not cleared, environment not started.

สารบัญ system

.kasidit/ directory per project:

.kasidit/
├── INDEX.md      # path → one-line purpose
├── RELATIONS.md  # module relation diagram
└── MEMORY.md     # user-confirmed project facts

• INDEX.md stores paths only — path/to/file.md — one-line purpose
• RELATIONS.md shows how modules connect; auto-detected, user-approved
• MEMORY.md accumulates verified facts across sessions

On error recurrence, look up RELATIONS.md before re-debugging. Standard markdown — linkable, greppable.

Point-based output with lazy loading

AI replies in points. Each point references a .kasidit/knowledge/<id>.md if deeper context exists. AI does not load the .md into its own context — the user clicks to read. Keeps agent context lean; makes knowledge auditable.

What changed vs (there is no prior)

This is the first release.

Breaking changes

Not applicable.

Migration

First-time setup — scaffold .kasidit/ per project (automated from v0.9.2 via /kasi-init).

See also

• v0.2.0 — UI Override Mode (next release)
• v0.3.0 — tier adaptation (where Rule 11 is enforced)
• v0.9.2 — /kasi-init automates the sarabanya scaffold