docs: answer-first README opening + anchor doc (why raw exports fail as agent context) - #1
Conversation
…l as agent context - docs/why-note-exports-fail-as-agent-context.md (new): four measured failure modes (empty shells, missing/lying dates, link dialects, unreviewed compression) -> six criteria a compile step must satisfy -> how kb-init implements each -> honest limitations - README.md: problem-first opening line + pointer to the anchor doc - README.zh-CN.md: same opening change + one-line Chinese pointer to the anchor doc (no parallel translation)
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 67cb0d8ba2
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| - **Link invariant:** output paths are frozen before link rewriting; unresolvable links | ||
| degrade to plain text and are recorded in `manifest.json`'s `unresolved_links`; under |
There was a problem hiding this comment.
Qualify link degradation in wikilink mode
When users select --wikilinks, unresolved non-ambiguous wikilinks do not degrade to plain text: _convert_links preserves them, and README.md:309-312 explicitly documents that output may contain links to nonexistent files. This unconditional implementation claim therefore gives users of that supported mode a false link-safety guarantee; limit it to the default mode or describe the wikilink exception.
Useful? React with 👍 / 👎.
| each one. All numbers below are measured on two real exports and reproduced in the | ||
| [README](../README.md#what-it-actually-does-in-numbers). |
There was a problem hiding this comment.
Correct the measurement provenance
Revise this provenance statement because not all subsequent numbers come from the two exports or are reproduced in the README: the 0% mtime result at lines 20-21 was measured on a separate maintained knowledge base, and the 11–15-level nesting result at lines 25-26 is sourced from docs/DESIGN.md rather than the README. As written, the anchor document misstates the evidence base it is intended to make auditable.
Useful? React with 👍 / 👎.
Files changed
docs/why-note-exports-fail-as-agent-context.md(new): the anchor document. Four measured failure modes of pointing an agent at a raw note export (empty shells ~54–61%, missing/lying dates — 5.2%/6.3% parseable creation dates, 0% truthful mtime staleness; link dialects that don't survive flattening; unreviewed compression) → six criteria a compile step must satisfy (drop shells but keep receipts; never emit a live wrong link; honest date fallback with a coverage gate; human gate on archive content; state its own coverage; land in the file the agent reads) → how kb-init implements each → honest limitations.README.md: opening line changed to problem-first ("A years-old note export is mostly empty shells — an AI agent can't use it as-is…") + one pointer line to the anchor doc. No forced question phrasing.README.zh-CN.md: same opening change + one-line Chinese pointer to the anchor doc (no parallel translation).Evidence base
README.mdanddocs/DESIGN.md; nothing new is invented. Internal-only DESIGN content (naming deliberations, competitor analysis) is excluded from the public doc.Also in this PR (metadata, reversible)
gh repo edit(done separately as a repo setting, not a file in this branch).Not in this PR
Homepage change deferred until after merge (it points at the new anchor doc).