egorfedorov

cco-config

View and tune CCO's behavior thresholds — re-read warnings, cache staleness, prompt-coach length bands, and the /cco-pack budget cap

egorfedorov 102 10 Updated 4w ago
GitHub

Install

npx skillscat add egorfedorov/claude-context-optimizer/cco-config

Install via the SkillsCat registry.

About this skill

We need to produce a 2-3 sentence plain-text summary, objective, factual, no marketing, no superlatives, no calls to action. Must be at most 60 words. No quotes, no markdown, no bullet points, no headings. Just plain text. Summarize: skill allows viewing and tuning CCO's behavior thresholds: re-read warnings, cache staleness, prompt-coach length bands, /cco-pack budget cap. It solves problem of magic numbers scattered across modules, providing centralized config with validation. Use when developer wants to adjust optimizer behavior, view current settings, or reset to defaults.

SKILL.md

CCO Config — tune how aggressive the optimizer is

CCO's behavior knobs used to be magic numbers spread across four modules.
This surfaces them, range-validated, in ~/.claude-context-optimizer/config.json
under thresholds.

The plugin root is the directory containing src/config.js. Run everything
from there.

Parse $ARGUMENTS:

show (or no arguments)

node src/config.js

Print the table as-is. * marks values the user set; ! marks a stored value
that failed validation and is being ignored in favour of the default. If any
! rows appear, point them out — a typo there means a knob silently isn't
doing what the user thinks.

get <key>

node src/config.js get <key>

set <key> <value>

node src/config.js set <key> <value>

Exits non-zero with the valid range on a bad value or unknown key — relay that
message rather than guessing a correction. After a successful set, say which
behavior it changes (see the table below), because the effect is not obvious
from the key name alone.

reset [key]

node src/config.js reset <key>    # one key
node src/config.js reset          # everything back to defaults

Resetting everything discards all the user's tuning — confirm first unless they
were explicit ("reset all", "вернуть всё по умолчанию").

The keys

Key Default What it changes
rereadWarnAt 3 reads of an unedited file before the first re-read warning
rereadEscalateAt 5 reads before the "put it in CLAUDE.md" escalation
bigFileLines 500 full-read line count that triggers the offset/limit warning
mediumFileLines 200 full-read line count that triggers the soft hint
staleTokenRatio 0.10 share of budget loaded after a file before Read Cache re-allows it
staleFiles 8 other files loaded after a file before it counts as evicted (at 200K)
staleTimeMs 600000 ms since last read before a cached file counts as stale
promptMinWords 8 word count below which the coach calls a prompt too vague
promptIdealMaxWords 200 upper bound of the coach's "ideal length" band
promptTooLongWords 500 word count above which the ask is considered buried
packBudgetPercent 25 max share of the budget /cco-pack may consume

Tuning guidance

  • Too many re-read warnings? Raise rereadWarnAt to 4–5.
  • Read Cache blocking files you still need? Lower staleTimeMs (e.g.
    300000 = 5 min) or staleTokenRatio so cached entries go stale sooner.
  • Read Cache not saving much? Raise them — entries stay fresh longer and
    more repeat reads get blocked.
  • Working on a 1M-context model with long prompts? Raise
    promptTooLongWords; the default band is tuned for shorter asks.
  • CCO_STALE_TIME_MS in the environment still overrides staleTimeMs, for a
    one-off experiment without changing config.

Categories