Policy-Driven Timeout and Limit Configuration System with Adaptive Overrides

[github-actions]

Strategic Improvement

Platform has 48 hardcoded values and 67 fallback patterns, limiting ability to self-optimize. Build policy.json system for all timeouts, iteration limits, thresholds with adaptive override mechanism based on runtime conditions (issue complexity, historical data, budget). Moves platform toward AGI-level self-tuning.

Acceptance Criteria

• Extracts all hardcoded timeouts, limits, thresholds from scripts to config/policy.json with: default, min, max, adaptive_strategy
• Implements adaptive override system: reads policy, applies context-specific adjustments (issue complexity score, historical stage duration, remaining budget)
• Adds helper function _smart_policy(category, key) in lib/compat.sh for script access
• Migrates at least 30 hardcoded values across: loop iteration limits, stage timeouts, retry counts, circuit breaker thresholds
• Documents policy schema and adaptive strategies in docs/POLICY.md
• Reduces platform hygiene hardcoded count from 48 to <20

Context

• Priority: P6
• Complexity: full
• Generated by: Strategic Intelligence Agent
• Strategy alignment: P6: Platform Self-Improvement (AGI-Level Readiness)

[github-actions]

⚓ Shipwright Pipeline Starting

Field	Value
Template	standard
Complexity	full
Max Iterations	30
Triage	Centralizes 48+ hardcoded timeouts/limits into adaptive policy system; strategic platform self-improvement extending existing config patterns (daemon-config.json, effort_levels) already referenced in CLAUDE.md.
Runner	ubuntu-latest
Risk Level	unknown
Workflow	24294833382

The autonomous pipeline is processing this issue. Updates will be posted as stages complete.

[github-actions]

[github-actions]

[github-actions]

[github-actions]

📋 Implementation Plan

Click to expand full plan (21 lines)

The plan is written (215 lines). Here's a summary of the implementation plan:

Plan Summary

Goal: Extract 48+ hardcoded timeouts, limits, and thresholds into a centralized config/tunables.json with adaptive overrides, reducing hardcoded count from 48 to <20.

Key components:

1. config/tunables.json (new) — 63 values across 11 categories (loop, daemon, health, adaptive, failure, predictive, decompose, convergence, pipeline_quality, context, helpers), each with default/min/max/adaptive_strategy

2. _smart_policy(category, key, default) in lib/compat.sh — New helper function with precedence chain: env var SW_<CAT>_<KEY> -> daemon-config.json -> tunables.json (with min/max clamping) -> hardcoded default. Same safety patterns as existing _smart_int().

3. Migration across 16 files — sw-loop.sh (9), sw-daemon.sh (12), sw-adaptive.sh (8), 6 lib/ files (19), 3 script files (9) = 57 total values migrated

4. Schema + docs — Updated policy.schema.json and docs/config-policy.md with tunables documentation and adaptive strategy explanations

Design choice: New _smart_policy() function rather than extending _smart_int(), because the min/max clamping and adaptive_strategy metadata are genuinely new capabilities. Existing helpers (_smart_int, policy_get, _config_get) remain untouched.

Critical risk mitigated: Default value mismatch (silent regression) — every tunables.json default is set to the exact current hardcoded value, verified by full test suite and grep audit.

11 discrete tasks with clear dependencies and testing at each step.

---

Generated by shipwright pipeline at 2026-04-12T00:40:05Z

[github-actions]

[github-actions]

[github-actions]

📐 Architecture Decision Record

Click to expand ADR (295 lines)

ADR written to .claude/pipeline-artifacts/design.md. Here's what the design covers:

Core decision: New _smart_policy(category, key, default) function in scripts/lib/compat.sh + new config/tunables.json file with 63 bounded tunables across 11 categories.

Key architectural choices:

• Separate function from _smart_int() to avoid overloading with clamping logic
• 4-layer precedence: env var -> daemon-config -> tunables.json (with min/max clamping) -> hardcoded default
• Env/daemon-config overrides are NOT clamped (deliberate operator choice); clamping only on tunables.json defaults
• adaptive_strategy is metadata-only, consumed by self-optimizer, not by _smart_policy() itself

4 alternatives rejected: extending _smart_int() (footgun), unifying on _config_get() (blast radius), merging into defaults.json (schema coupling), merging into policy.json (mixed concerns)

Scope: 1 new file, 18 modified files, 63 values migrated, zero new dependencies, zero breaking changes
errides for model routing, effort levels, loop tuning. Read via _smart_int() / _smart_model() / _smart_effort() in scripts/lib/compat.sh.

Gap: None of these layers carry min/max bounds or adaptive strategy metadata. An operator setting loop.extension_size=0 or the self-optimizer pushing circuit_breaker_threshold=100 would silently break pipeline behavior. The 48 hardcoded values have no central registry and no safe override path.

Constraints:

• All scripts must remain Bash 3.2 compatible (no associative arrays, no ${var,,})
• Existing callers of _smart_int(), _smart_model(), _smart_effort(), policy_get(), _config_get() must not change behavior
• Default values must exactly match current hardcoded values (zero behavioral regression)
• jq is the only JSON tool available; called at script init, never in hot loops

Decision

Introduce a new configuration file config/tunables.json and a new helper function _smart_policy() in scripts/lib/compat.sh. This is an additive change that creates a fourth config tier specifically for operational tunables with safety bounds.

Pattern: Bounded Tunable with Adaptive Metadata

Each tunable is a JSON object with four fields:

{
  "loop": {
    "extension_size": {
      "default": 5,
      "min": 1,
      "max": 20,
      "adaptive_strategy": "complexity_scaled"
    }
  }
}

• default -- exact current hardcoded value (verified by grep audit)
• min / max -- safety bounds enforced at read time (clamped, not rejected)
• adaptive_strategy -- declarative hint for the self-optimizer ("fixed", "complexity_scaled", "budget_aware", "load_sensitive", "history_based")

Function: _smart_policy(category, key, default)

# _smart_policy <category> <key> [default]
# Precedence chain (first non-empty wins):
#   1. SW_<CATEGORY>_<KEY> env var (e.g., SW_LOOP_EXTENSION_SIZE)
#   2. daemon-config.json .<category>.<key>
#   3. tunables.json .<category>.<key>.default (with min/max clamping)
#   4. Hardcoded default (third argument)
#
# Clamping: if tunables.json defines min/max for this key,
#           the resolved value is clamped: max(min, min(max, value))

</details>

---
_Generated by `shipwright pipeline` design stage at 2026-04-12T00:48:42Z_

[github-actions]

[github-actions]

🔨 Build started — shipwright loop with 30 max iterations, 1 agent(s), model: opus

[github-actions]