Skip to content

docs: correct five claims before 0.15.0 - #96

Draft
BitmapAsset wants to merge 9 commits into
fix/explain-header-hardeningfrom
docs/launch-truth-pass
Draft

BitmapAsset wants to merge 9 commits into
fix/explain-header-hardeningfrom
docs/launch-truth-pass

Conversation

@BitmapAsset

Copy link
Copy Markdown
Owner

Five places where the repository's text said more than the code does, or disagreed with another page. Wording only; no behaviour or output format changes.

Stacked on #95 (fix/explain-header-hardening).

  • docs/receipts.md: the receipt no longer claims "a chain that detects any edit". It now says what probbit live verify does: an edit to a line of the strand is found (the replay names the earliest line that differs); a log rewritten from its first line is not, unless its head digest was kept elsewhere. This matches docs/persona.md and docs/soul-to-habits.md.
  • docs/adapters.md: the MCP hosts row claimed real calls from Claude Code and Codex CLI. docs/agents.md records real calls from Claude Code only; for Codex CLI it records the config entry the CLI writes, with no model call. The row now says the same.
  • README.md: "npm may print an allow-scripts warning" is now "npm may print a warning about install scripts", which holds for both spellings current npm releases use.
  • probbit monitor: its help text, the man page, docs/persona.md §5.8, docs/monitor-app.md and the OpenClaw plugin README now say "internal state" where they said "inner state", in line with docs/welfare-stance.md. The §5.8 heading anchor changed; the old anchor #58-monitor-watch-an-individuals-inner-state is kept with an <a id> so existing links still resolve. One line under Unreleased in CHANGELOG.md.
  • docs/probbit-flow.svg and docs/probbit-flow.png: the figure no longer says "zero dependencies" (the default build has one third-party crate, libm, README). The PNG was re-rendered at its original 3200 x 2000; it differs from the old one only in that text line.

docs/probbit.1 was regenerated with probbit man and llms-full.txt with its generator.

Validation on the host (macOS arm64), from a clean build, run before the last rebase (which changed only plugin
TypeScript and markdown below this branch):

With SHELL unset or empty the hint named only ~/.profile, which an
interactive bash does not read; an outside newcomer run on Linux x86_64
showed it. It now names ~/.zshrc on macOS and "~/.bashrc (bash) or
~/.profile" elsewhere. A set SHELL is unchanged. The installer test
checks the empty and the unset case.
…imed

SECURITY.md, the CLI's and probbit-ir's crate documentation and the
pull request checklist still said there are no external crates or that
the shipped path builds offline. An outside newcomer run on Linux x86_64
showed a from-source build fetching libm. They now say that the default
build has one third-party crate, libm (the nostd build adds hashbrown
and portable-atomic), and SECURITY.md says that probbit monitor listens
on 127.0.0.1 and that probbit opens no outbound connection.
… line

An outside newcomer run on Linux x86_64 showed three statements that did
not hold there:
- the binary is 3 to 5 MB, not 3 to 4 MB (the 0.14.0 release binaries
  measure 3.7 to 5.0 MB; README and npm README);
- the 60-second tour's sixty seconds are the demo with a built binary;
  the first build takes a few minutes and the whole test suite several
  more. The --hard runs keep their Apple M4 results; on a slower machine
  raise --budget-ms until partial and then diagnostics_passed appear;
- probbit persona init prints its line on stderr (README, AGENTS.md and
  npm README quick starts).
llms-full.txt regenerated.
The closure in the curiosity test captures nothing mutably; an outside
newcomer run on Linux x86_64 got an unused_mut warning from it.
The previous commit's range, 3 to 5 MB, had no room: the largest 0.14.0
release binary is 4.95 MB, and an outside from-source build on Linux
x86_64 measured 5.2 MB. README and npm README now say 3 to 6 MB. The
changelog entry on the crate statements now says what each file says:
probbit-ir's own default build has no third-party crate.
llms-full.txt regenerated.
… reports

bash, and macOS /bin/sh with it, sets SHELL to the account's login shell
when it starts with SHELL unset; dash and zsh leave it unset. The case
expected ~/.zshrc on macOS whatever the shell under test was, so it
passed where the login shell is zsh and failed on a CI runner whose
login shell is bash (the installer, rightly, named ~/.bash_profile).
It now asks the shell under test what it reports and expects the file
for that value, with the same mapping as the other cases, and prints
the value in its ok line. The changelog entry says when the new hint
applies.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant