Skip to content

Release notes: httpd plan autonomous loop

Summary

Adds scripts/httpd-plan-loop.py and goal-directed loop (code_implementer + per-todo goal file) so Phase H httpd work can run in a loop without manual nudges every 30 minutes.

Agent continuation

  1. Read docs/ecosystem/httpd-m1-baseline.md and plan YAML todos in docs/superpowers/plans/2026-05-16-li-httpd-plan.md.
  2. export CURSOR_API_KEY=... and LI_CURSOR_AGENTS_ROOT=../li-cursor-agents; cd lic && ./scripts/httpd-plan-loop.py --once.
  3. Review PR from agent; run full CI; merge when green.
  4. Blocked: Li routing binaries exit non-zero until CallProc string-to-extern ABI is fixed — gates use HTTPD_SKIP_LI_ROUTING_BIN=1 until then.

Changed

Path Change
scripts/httpd-plan-loop.py Parse plan todos, run gates, invoke code_implementer with todo goal via li-cursor-agents
scripts/httpd-plan-gates.sh Build + run_httpd_config.sh subset
docs/ecosystem/httpd-m1-baseline.md Merged/stale PR record
.cursor/automations/httpd-plan-loop.md Automation doc
packages/li-http/src/lib.li li_rt_str_prefix_is_get, single-call parse_request (E0311)
runtime/li_rt.c, li_rt.h C helpers for GET probe + parse tag
li-tests/run_httpd_config.sh HTTPD_SKIP_LI_ROUTING_BIN for compile-only gate
li-cursor-agents run-agent --goal-file + scripts/goal-directed-loop.sh (LI_REPO_WORKFLOW_REPO=lic)

Not changed

  • Full M1 plan completion (loop drives todos incrementally).
  • w0-lean-gate / async reactor (deferred by pick order).
  • Benchmark matrix refresh (agent must run per httpd rule after runtime edits).

Breaking

N/A — dev tooling and test gate flags only.

Security

N/A — no new trusted surface beyond existing li_rt_str_* helpers (audited in trusted-c-audit.toml).

Performance

N/A — loop is offline orchestration.

Downstream

  • benchmarks: scripts/httpd-plan-loop.sh wrapper.
  • li-cursor-agents: rebuild npm run build after pulling registry change.