keelrun-cli 0.4.0

The `keel` binary: run | init | doctor | status | explain. The product's face — every command has a byte-deterministic `--json` twin and stable exit codes (dx-spec §1–2, §5–6).
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25

# Keel smart-defaults pack — contracts-v1.
# This document ships compiled into the binary and is what Level 0
# (`keel run` with no keel.toml) applies to every discovered call site.
# A user keel.toml is layered on top of these defaults; any layer a user
# sets for a target replaces the corresponding default layer wholesale.
#
# Level 0 hard rules (DX spec §1) that are BEHAVIOR, not config, and cannot
# be expressed or overridden here:
#   - never change program semantics on the success path
#   - never retry non-idempotent calls (POST without idempotency key);
#     they are observed, not retried (KEEL-E014) and reported in `keel status`
#   - if a call site cannot be wrapped safely, do nothing and note it

[defaults.outbound]                 # any intercepted network call
timeout = "30s"
retry   = { attempts = 3, schedule = "exp(200ms, x2, max 30s, jitter)", on = ["conn", "timeout", "429", "5xx"] }
breaker = { failures = 5, cooldown = "15s" }

[defaults.llm]                      # any llm:* target; overrides outbound field-wise per layer
timeout = "120s"
retry   = { attempts = 6, schedule = "exp(500ms, x2, max 60s, jitter)", on = ["conn", "timeout", "429", "5xx"] }
# Provider-aware backoff: a Retry-After (or provider equivalent) on the error
# overrides the schedule wait: wait = max(schedule_wait, retry_after).
breaker = { failures = 5, cooldown = "30s" }
cache   = { mode = "dev" }          # dev-loop response cache; inert when KEEL_ENV=prod