agentic-loop

A command for a staged workflow that helps develop features and investigate operational problems in service repositories.

In plain words
What is it for?
It is for running a selected workflow profile against a target, inspecting repository health, planning changes, and—after approval—moving through later implementation phases.
Why use it?
It gives the work defined planning and approval points, so implementation does not begin automatically before the proposed steps are reviewed.

Command for Cursor

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add commands/pallerana/agentic-loop-engineering-kit/agentic-loop
Clone the repo
git clone --depth 1 https://github.com/pallerana/agentic-loop-engineering-kit

Made for: Cursor.

Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 2,004 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00000 $0.02004
Opus 5 $0.00000 $0.01002
Sonnet 5 $0.00000 $0.00401
Haiku 4.5 $0.00000 $0.00200

Measured yesterday against content hash 7cda34291241, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

agentic-loop scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.cursor/commands/agentic-loop.md · 157 lines

How it starts

The opening of the file, as written. The whole thing — 157 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Agentic Loop Engineering Kit

Modular Loop Engineering OS for your-service-* repos — feature development and ops troubleshooting.

Parse arguments

If the user invokes /agentic-loop with no arguments or --help only → print the HELP section below and do not start a loop.

Otherwise → SwitchMode(plan) then read and follow .cursor/skills/agentic-loop/SKILL.md (runbook: agentic-loop-runbook) with parsed flags and target.

PLAN MODE REQUIRED

  • Loop always starts in Plan mode — never autonomous Agent execution
  • Default mode: L1 (Phase 0→2, report-only, no code/git)
  • Stop after Phase 2 until user explicitly approves implementation
  • --mode L2 / L3-push require flag and user confirmation in chat before Phase 3+
  • Failure Router recommends next phase — does not auto-re-dispatch

HELP

Agentic Loop Engineering Kit — modular Loop Engineering OS for your-service-* repos

USAGE
  /agentic-loop --help
  /agentic-loop --profile <id> [options] [target]
   [options] [target]          # alias: --profile cell-health

PROFILES
  cell-health              Cell Health MVP (aggregator + central); extends default
  springboot-default   Any cellcp Java/Spring Boot repo (Gradle)
  ops-incident             Datadog / PagerDuty investigation (L1 default)
  cellassignments          cellcp assignments API (extends default)
  cellrouter               cellcp http cell router (extends default)
  cellmigrator             cellcp cell migrator (extends default)
  celldiscovery            cellcp cell discovery (extends default)
  cellpersistence          ODP cell persistence API (extends default)

MODES (maturity)
  L1   Report only — Phase 0→2, STATE/run-log, no code changes (DEFAULT)
  L2   Implement + review; PR draft; human may push (requires approval after Phase 2)
  L3-push  L2 + squash, push feature branch, babysit CI (max 3); human merge
       Explicit --mode L2|L3-push required; human must approve Phase 3+ and ship

OPTIONS
  --profile <id>           Required except for --help and cell-health alias
  --mode L1|L2|L3-push     Override profile default maturity
  --repo <path>            Repo dir under workspace (required for default profile)
  --handoff <profile>      Ops→feature: preload STATE from ops-incident run
  --from-state <path>      Resume from loop-kit/*-state.md
  --phase <n|name>         Start at phase (orchestrator validates prerequisites)
  --dry-run                Print planned phases + gates; no MCP/git writes
  --help                   This reference

TARGET (one of; profile-specific)
  <JIRA-KEY>               e.g. PROJ-153 (feature profiles)
  --datadog-monitor <id>   Ops: Datadog monitor id or query
  --pagerduty <id>         Ops: PagerDuty incident id (INC-…)
  --pr <n>                 PR review (Phase 5) or ship/hygiene (Phase 6/7 with --phase)

PR REVIEW (--pr or GitHub URL → Phase 5 @pr-code-review)
  --pr <n> alone           Phase 5 full review; repo from --profile / --repo
  <PR URL>                 Parse owner/repo/number; infer profile when possible
  --pr <n> + URL           URL wins if numbers differ; warn on conflict
  Default --mode L1        Report-only findings table (no auto-fix)
  pr_type auto-detect      java-only | infra-only | mixed → routed Pass 3 + drill
  Dedup                    Phase 0f registry suppresses repeat of open threads

PHASES (feature profiles; ops-incident uses subset)
  0    Context sync (wiki, pattern_doc, loop_playbook, Jira, graphify, profile skill)
  0b   Ticket ack (Jira In Progress / PagerDuty note)
  1    Plan or ops hypothesis (ce-plan)
  2    Plan / hypothesis review (incl. ce-testing-reviewer)
  3    Implement (ce-work)
  4    Verify (gradle, JaCoCo, diff coverage, E2E tier)
  5    Code review (@pr-code-review when --pr or PR URL set)
  6    Ship (squash, push, PR, babysit)
  7    PR hygiene (Copilot/CodeQL threads — not full review)
  8    Close-the-loop (Jira comment + PR + tests + SHA)
  9a   Wiki + graphify update
  9c   ce-compound (human wiki only)
  9d   Self-improvement — TOON contract + apply-loop-learning (L2/L3 only)

PHASE 9d (self-improvement) — see .cursor/agents/agentic-loop-orchestrator.md (SSOT)

<!-- PHASE-9D-TRANSITIONS:verbatim:start — do not edit in mirrors -->
| Step | Action | Transition |
|------|--------|------------|
| 9a | wiki log + current-status | always |
| 9b | graphify update | always |
| 9c | ce-compound | always |
| 9d-eligible | check L2/L3 + `self_improvement` | skip → end |
| 9d-extract | loop-self-improvement | `NO_LEARNINGS` → end |
| 9d-preflight | `scripts/loop_9d_preflight.sh` | fail closed; run before validate **and** before apply |
| 9d-validate | apply script, no approve | `PENDING_APPROVAL` → STOP |
| 9d-gate | present summary table | **STOP — same-turn approve forbidden** |
| 9d-apply | user approve → `approve` | `APPLY_SUCCESS` / `IDEMPOTENT_SKIP` |
| 9d-promo | promotion contract if any | after learning applied |
<!-- PHASE-9D-TRANSITIONS:verbatim:end -->

  L1 skips 9d
  Validate: apply-loop-learning.sh --contract (no approve) → PENDING_APPROVAL
  --dry-run → PENDING_APPROVAL (not APPLY_SUCCESS)
  Human STOP after staging summary; approve only on user message
  See loop-kit/contracts/README.md

FAILURE ROUTER (orchestrator; not a separate agent)
  delta/JaCoCo fail     → test-repair pass → implementer
  local CI fail         → implementer (then planner if 2+ arch failures)
  remote CI fail        → ship/babysit (max 3) → planner with logs
  AC/plan gap           → planner
  ops infra/flake       → human inbox in STATE.md

EXAMPLES — Feature (Cell Health)
  /agentic-loop --help
  /agentic-loop --profile cell-health PROJ-153
  /agentic-loop --profile cell-health --mode L1 PROJ-156
   PROJ-153

EXAMPLES — Feature (other cellcp repo)
  /agentic-loop --profile cellassignments PROJ-999
  /agentic-loop --profile springboot-default \
    --repo your-service-api PROJ-999

EXAMPLES — Ops (Datadog / PagerDuty)
  /agentic-loop --profile ops-incident --pagerduty INC-ABC123
  /agentic-loop --profile ops-incident --datadog-monitor 12345678

EXAMPLES — Handoff (ops → fix)
  /agentic-loop --profile ops-incident --handoff cell-health \
    --repo your-service-cell-health-aggregator PROJ-153 \
    --from-state loop-kit/ops-incident-state.md

EXAMPLES — Resume / ship only
  /agentic-loop --profile cell-health --phase 6 --pr 2

EXAMPLES — PR review (Phase 5, report-only)
  /agentic-loop --profile cellrouter --mode L1 --pr 78
  /agentic-loop --profile cellassignments --mode L1 \
    https://github.com/<org>/your-service-api/pull/123
  /agentic-loop --profile cell-health --phase 5 --pr 2
  /agentic-loop --profile springboot-default --repo infra/ --mode L1 --pr <n>

EXAMPLES — PR thread hygiene only (Phase 7, after green CI)
  /agentic-loop --profile cell-health --phase 7 --pr 2

FILES
  Skill       .cursor/skills/agentic-loop/SKILL.md
  PR review   .cursor/rules/pr-code-review.mdc
  Profiles    loop-kit/profiles/*.yaml
  State       loop-kit/{profile}-state.md, STATE.md, loop-run-log.md
  Gates       loop-kit/GATES.md

SEE ALSO
  pr-code-review.mdc, cell-health-mvp skill, service-quality-drill.mdc, docs/wiki/your-loop-wiki.md

Read the full file on GitHub · 157 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. yesterday First seen · 157 lines · 0 tokens per session scan A 7cda34291241

Subscribe to this mod's changes

agentic-loop is a command published in the GitHub repository pallerana/agentic-loop-engineering-kit (4 stars, last pushed 2mo ago), licensed Apache-2.0. It costs nothing until one of its globs matches a file; then it loads 2,004 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.