Release v1.3.2: A reasoning baseline that does not depend on remembering a keyword · fidpa/cc-telegram-bot

A reasoning level can now be set once instead of per message. THINKING_DEFAULT
(off by default = previous behaviour) is the baseline thinking level for every message,
applied before keywords. Until now the only way to get reasoning effort was a keyword in
the message, so a bot whose turns are judgement calls rather than commands depended on the
operator remembering one every time, which is the kind of discipline that fails on the
message where it mattered. normal maps to effort: high, deep to effort: max.
Keywords still apply on top and can only raise the level, never lower it: a deep baseline
stays deep even when the message contains a normal keyword, since a floor that wording
can undo is not a floor

/model o now selects Opus 5 instead of Opus 4.8. Opus 4.8 stays in the display-name
table so an existing CLAUDE_MODEL pointing at it is still shown by name rather than as a
raw ID

Two lines in .env.example stopped the bot from starting as a LaunchAgent. They broke
source .env, which is how launchagent/start.sh reads the file. Uncommenting them and
starting the bot produced command not found: are and command not found: the instead of
a running bot. Every other multi-word value in the file was already quoted; these two were
not:
- the BOT_PERSONA example, added in 1.2.3, now quoted, with a note saying why
- a prose line in the ALLOWED_TOOLS section that happened to wrap onto
  # ALLOWED_TOOLS=Read,Grep,Glob the built-ins reduce..., so a sentence read as an
  assignment. Reflowed. It also looked like a variable to the CI documentation gate, which
  greps ^#? ?[A-Z_]{4,}=; it passed only because ALLOWED_TOOLS is a real variable, and
  a prose line starting with any other uppercase token would have failed the gate with a
  confusing message
Found while configuring a bot built on this code, not by review: the failure surfaces on
the LaunchAgent path and not when running bun run start by hand

Navigation Menu