Agent Planning: Workflows, HTN, Plan-Execute · Ingénierie IA

Pick a workflow, an HTN or evolutionary planner, or a replanning executor. Stdlib Python from the lessons.

Pick a planner

Factor	Workflow	HTN + ChatHTN	AlphaEvolve	ReAct / ReWOO
Who decides	Engineers	Symbolic layer	Evaluator selects	The model
Guarantee	Bounded step count	Provable soundness	Best plan by fitness	No formal guarantees
Needs	Steps you can enumerate	Operator schema	Machine-checkable fitness	LLM in the loop
Fits	Predictable, compliance-bound	Scheduling, policy-bound	Compiler optimization	Open-ended research
LLM role	Calls on fixed code paths	Fallback decomposition	Proposes mutations	Picks each step

Sources

Five workflow patterns in code

Task	Code
Prompt chaining: output N is input N+1	`prompt_chain(input_text, llm, steps)`
Routing: classify, then dispatch	`route(input_text, classifier, handlers)`
Parallelization: N runs, majority vote	`parallel_vote(prompt, llm, n=5)`
Orchestrator-workers: pick specialists	`orchestrator_workers(task, workers, synth)`
Evaluator-optimizer: loop until PASS	`evaluator_optimizer(task, proposer, evaluator, max_iter=5)`
Unknown route label	`handlers.get("default")` gives "escalate to human"
Run all five demos	`python3 code/main.py`

Sources

Plan-execute executor

Planner planner(goal, history, None) returns the first plan. Event plan.commit.
Executor runs plan[cursor] through the dispatcher. Events step.start and step.end.
Executor on success, append to history and move the cursor to the next step.
Executor on an exception, store step.error and set last_error.
Budget check replans_used >= self.max_replans returns failed with replan_budget.
Planner planner(goal, history, last_error) gives a new plan. Event plan.draft.
Executor emits plan.diff, commits the new plan and sets cursor = 0.
Executor the last step succeeds. Event session.complete with goal_met.

Sources

Evidence plan waves

Nodes and connections

Node	Role	Kind
contract	wave 1, review contract	process
implementation	wave 2, unittest	process
docs	wave 2, check_links.py	process
integration	wave 3, acceptance gate	process

From	To	Connection
contract	implementation	depends_on
contract	docs	depends_on
implementation	integration
docs	integration

Sources

Run a replanning agent

agent = PlanExecuteAgent(

planner=make_deterministic_planner(fail_step_id=2, recovery="route_around"),

executor=executor,(tool_name, args) -> result, raises ToolFailure

max_steps=12, max_replans=5,both budgets are hard

)

res = agent.run("ship the report")

res.status, res.reason"completed" and "goal_met"

res.revisionslist of PlanDiff

r.to_dict()revision, removed, added, revised

Step(2, "transform", {"mode": "v2"}, "computed via fallback")

Sources

phases/19-capstone-projects/24-plan-execute-control-flow/code/main.py

HTN planner with LLM fallback

op = Operator("open_editor", ("logged_in",), ("editor_open",))

m = Method("ship_change_m1", "ship_change", ("logged_in",),

("open_editor", "write_tests", "run_tests", "open_pr"))

planner = HTNPlanner(operators=operators, methods=methods, llm=llm)

plan = planner.plan("ship_change", {"logged_in"})list of operators or None

planner.cached_methodsLLM decompositions kept for the next call

planner.llm.calls0 new calls on a cache hit

Sources

phases/14-agent-engineering/11-planning-htn-and-evolutionary/code/main.py

Planning terms

Operator: primitive action with preconditions and effects.
Method: a way to break a compound task into subtasks, with preconditions.
ChatHTN: the symbolic planner asks the LLM when no method matches.
Fitness function: deterministic, machine-checkable score over outputs.
Augmented LLM: LLM plus search, tools and memory. The atomic unit.
expected_outcome: planner's success condition. The executor does not enforce it.
Replan: hand the error to the planner. This turns a script into an agent.

Sources

Defaults

Setting	Default	Note
`max_steps`	`12`	All step executions, replans included
`max_replans`	`5`	Planner calls after the first plan
`max_depth`	`12`	HTN recursion limit, then None
`max_iter`	`5`	Evaluator-optimizer rounds
`n`	`5`	Voters in `parallel_vote`
`generations`	`12`	Evolutionary demo, keep top 3 of 6

Sources

Failures and fixes

Symptom	Cause	Fix
`failed`, reason `replan_budget`	Planner returns the same broken plan	Feed `last_error` and route around it
`failed`, reason `step_budget`	Replans add steps past `max_steps`	Shorter plans or reuse history
`failed`, reason `no_plan`	Planner returned `[]`	Add a recovery plan for that error
`dependency cycle among:`	Items wait on each other	Resolve the disagreement behind it
`has no repository evidence`	Item has empty `evidence`	Cite a repository fact
`has no proof command`	Empty `proof`	Give one exact check
HTN soundness collapses	No precondition/effect schemas	Validate LLM output against operators
Evolution does not converge	LLM asked if the code is better	Use a deterministic, fast evaluator

Sources

Commands

Command	What it does
`python3 code/main.py`	Runs the lesson demo and prints the trace
`python3 -m unittest discover code/tests -v`	Runs the evidence plan tests
`outputs/evidence-plan.json`	Written by lesson 44, status ready or blocked

Sources

Questions

Workflow or agent?: Workflow when you can enumerate the steps. Agent when the next step depends on the last result.
Why cap replans and steps?: A capped replan fails faster with a clearer reason.
Why is ChatHTN sound?: LLM output enters only as candidate decompositions. The schema rejects invalid ones.
Which steps run together?: Steps in the same wave. Integration waits for both branches.

Sources

Tips

Start with direct API calls. Reach for ReAct or ReWOO before HTN.
Define the contract before its implementations.
Give each work item an id, change, evidence, dependencies and proof.
Emit a plan diff on each revision. Do not rewrite the plan silently.
Store the plan next to the work, not in chat checkboxes.

Sources

Original source

---
id: agent-planning
title: Agent Planning: Workflows, HTN, Plan-Execute
code: AW
subtitle: Pick a workflow, an HTN or evolutionary planner, or a replanning executor. Stdlib Python from the lessons.
hue: 32
sources: phases/14-agent-engineering/11-planning-htn-and-evolutionary, phases/14-agent-engineering/12-anthropic-workflow-patterns, phases/19-capstone-projects/24-plan-execute-control-flow, phases/14-agent-engineering/44-plan-from-evidence
source_hash: 28f779dbd6acbd70
---

## Pick a planner
type: comparison
from: phases/14-agent-engineering/11-planning-htn-and-evolutionary/docs/en.md, phases/14-agent-engineering/12-anthropic-workflow-patterns/docs/en.md

| Factor | Workflow | HTN + ChatHTN | AlphaEvolve | ReAct / ReWOO |
|---|---|---|---|---|
| Who decides | Engineers | Symbolic layer | Evaluator selects | The model |
| Guarantee | Bounded step count | Provable soundness | Best plan by fitness | No formal guarantees |
| Needs | Steps you can enumerate | Operator schema | Machine-checkable fitness | LLM in the loop |
| Fits | Predictable, compliance-bound | Scheduling, policy-bound | Compiler optimization | Open-ended research |
| LLM role | Calls on fixed code paths | Fallback decomposition | Proposes mutations | Picks each step |

## Five workflow patterns in code
type: task-table
from: phases/14-agent-engineering/12-anthropic-workflow-patterns/code/main.py, phases/14-agent-engineering/12-anthropic-workflow-patterns/docs/en.md
span: 2

| Task | Code |
|---|---|
| Prompt chaining: output N is input N+1 | `prompt_chain(input_text, llm, steps)` |
| Routing: classify, then dispatch | `route(input_text, classifier, handlers)` |
| Parallelization: N runs, majority vote | `parallel_vote(prompt, llm, n=5)` |
| Orchestrator-workers: pick specialists | `orchestrator_workers(task, workers, synth)` |
| Evaluator-optimizer: loop until PASS | `evaluator_optimizer(task, proposer, evaluator, max_iter=5)` |
| Unknown route label | `handlers.get("default")` gives "escalate to human" |
| Run all five demos | `python3 code/main.py` |

## Plan-execute executor
type: flow
from: phases/19-capstone-projects/24-plan-execute-control-flow/docs/en.md, phases/19-capstone-projects/24-plan-execute-control-flow/code/main.py

1. **Planner**: `planner(goal, history, None)` returns the first plan. Event `plan.commit`.
2. **Executor**: runs `plan[cursor]` through the dispatcher. Events `step.start` and `step.end`.
3. **Executor**: on success, append to history and move the cursor to the next step.
4. **Executor**: on an exception, store `step.error` and set `last_error`.
5. **Budget check**: `replans_used >= self.max_replans` returns failed with `replan_budget`.
6. **Planner**: `planner(goal, history, last_error)` gives a new plan. Event `plan.draft`.
7. **Executor**: emits `plan.diff`, commits the new plan and sets `cursor = 0`.
8. **Executor**: the last step succeeds. Event `session.complete` with `goal_met`.

## Evidence plan waves
type: architecture
from: phases/14-agent-engineering/44-plan-from-evidence/docs/en.md, phases/14-agent-engineering/44-plan-from-evidence/code/main.py

```arch
node | contract | 1,1 | contract | wave 1, review contract | process
node | impl | 2,1 | implementation | wave 2, unittest | process
node | docs | 2,2 | docs | wave 2, check_links.py | process
node | integ | 3,1 | integration | wave 3, acceptance gate | process
edge | contract | impl | depends_on
edge | contract | docs | depends_on
edge | impl | integ |
edge | docs | integ |
```

## Run a replanning agent
type: annotated-code
lang: python
from: phases/19-capstone-projects/24-plan-execute-control-flow/code/main.py

```python
agent = PlanExecuteAgent(
    planner=make_deterministic_planner(fail_step_id=2, recovery="route_around"),
    executor=executor,              # (tool_name, args) -> result, raises ToolFailure
    max_steps=12, max_replans=5,    # both budgets are hard
)
res = agent.run("ship the report")
res.status, res.reason              # "completed" and "goal_met"
res.revisions                       # list of PlanDiff
r.to_dict()                         # revision, removed, added, revised
Step(2, "transform", {"mode": "v2"}, "computed via fallback")
```

## HTN planner with LLM fallback
type: annotated-code
lang: python
from: phases/14-agent-engineering/11-planning-htn-and-evolutionary/code/main.py

```python
op = Operator("open_editor", ("logged_in",), ("editor_open",))
m = Method("ship_change_m1", "ship_change", ("logged_in",),
           ("open_editor", "write_tests", "run_tests", "open_pr"))
planner = HTNPlanner(operators=operators, methods=methods, llm=llm)
plan = planner.plan("ship_change", {"logged_in"})  # list of operators or None
planner.cached_methods      # LLM decompositions kept for the next call
planner.llm.calls           # 0 new calls on a cache hit
```

## Planning terms
type: concepts
from: phases/14-agent-engineering/11-planning-htn-and-evolutionary/docs/en.md, phases/14-agent-engineering/12-anthropic-workflow-patterns/docs/en.md, phases/19-capstone-projects/24-plan-execute-control-flow/docs/en.md

- **Operator**: primitive action with preconditions and effects.
- **Method**: a way to break a compound task into subtasks, with preconditions.
- **ChatHTN**: the symbolic planner asks the LLM when no method matches.
- **Fitness function**: deterministic, machine-checkable score over outputs.
- **Augmented LLM**: LLM plus search, tools and memory. The atomic unit.
- **expected_outcome**: planner's success condition. The executor does not enforce it.
- **Replan**: hand the error to the planner. This turns a script into an agent.

## Defaults
type: defaults
from: phases/19-capstone-projects/24-plan-execute-control-flow/code/main.py, phases/19-capstone-projects/24-plan-execute-control-flow/docs/en.md, phases/14-agent-engineering/11-planning-htn-and-evolutionary/code/main.py, phases/14-agent-engineering/12-anthropic-workflow-patterns/code/main.py

| Setting | Default | Note |
|---|---|---|
| `max_steps` | `12` | All step executions, replans included |
| `max_replans` | `5` | Planner calls after the first plan |
| `max_depth` | `12` | HTN recursion limit, then None |
| `max_iter` | `5` | Evaluator-optimizer rounds |
| `n` | `5` | Voters in `parallel_vote` |
| `generations` | `12` | Evolutionary demo, keep top 3 of 6 |

## Failures and fixes
type: troubleshooting
from: phases/19-capstone-projects/24-plan-execute-control-flow/code/main.py, phases/19-capstone-projects/24-plan-execute-control-flow/docs/en.md, phases/14-agent-engineering/44-plan-from-evidence/code/main.py, phases/14-agent-engineering/11-planning-htn-and-evolutionary/docs/en.md

| Symptom | Cause | Fix |
|---|---|---|
| `failed`, reason `replan_budget` | Planner returns the same broken plan | Feed `last_error` and route around it |
| `failed`, reason `step_budget` | Replans add steps past `max_steps` | Shorter plans or reuse history |
| `failed`, reason `no_plan` | Planner returned `[]` | Add a recovery plan for that error |
| `dependency cycle among:` | Items wait on each other | Resolve the disagreement behind it |
| `has no repository evidence` | Item has empty `evidence` | Cite a repository fact |
| `has no proof command` | Empty `proof` | Give one exact check |
| HTN soundness collapses | No precondition/effect schemas | Validate LLM output against operators |
| Evolution does not converge | LLM asked if the code is better | Use a deterministic, fast evaluator |

## Commands
type: command-table
from: phases/14-agent-engineering/44-plan-from-evidence/docs/en.md, phases/19-capstone-projects/24-plan-execute-control-flow/code/main.py, phases/14-agent-engineering/11-planning-htn-and-evolutionary/docs/en.md

| Command | What it does |
|---|---|
| `python3 code/main.py` | Runs the lesson demo and prints the trace |
| `python3 -m unittest discover code/tests -v` | Runs the evidence plan tests |
| `outputs/evidence-plan.json` | Written by lesson 44, status ready or blocked |

## Questions
type: qa
from: phases/14-agent-engineering/12-anthropic-workflow-patterns/docs/en.md, phases/14-agent-engineering/11-planning-htn-and-evolutionary/docs/en.md, phases/19-capstone-projects/24-plan-execute-control-flow/docs/en.md, phases/14-agent-engineering/44-plan-from-evidence/docs/en.md

- **Workflow or agent?** Workflow when you can enumerate the steps. Agent when the next step depends on the last result.
- **Why cap replans and steps?** A capped replan fails faster with a clearer reason.
- **Why is ChatHTN sound?** LLM output enters only as candidate decompositions. The schema rejects invalid ones.
- **Which steps run together?** Steps in the same wave. Integration waits for both branches.

## Tips
type: tips
from: phases/14-agent-engineering/11-planning-htn-and-evolutionary/docs/en.md, phases/14-agent-engineering/12-anthropic-workflow-patterns/docs/en.md, phases/19-capstone-projects/24-plan-execute-control-flow/docs/en.md, phases/14-agent-engineering/44-plan-from-evidence/docs/en.md

- Start with direct API calls. Reach for ReAct or ReWOO before HTN.
- Define the contract before its implementations.
- Give each work item an id, change, evidence, dependencies and proof.
- Emit a plan diff on each revision. Do not rewrite the plan silently.
- Store the plan next to the work, not in chat checkboxes.