Coding Agents: Landscape and Capstones · Ingénierie IA

Pick a scaffold, set the budgets and gates, run the capstone agents and the three coding-agent projects.

JSON tool calls vs CodeAct

Factor	JSON tool calls	CodeAct
One action is	One structured JSON payload	A whole Python program
Audit	Easy, one action per turn	Harder, wide blast radius
Composition	Limited	Loops, chains tools, catches errors
Safety	Validator on every call	Needs a hardened sandbox
Toy blast radius	1 file per action	Every failing file at once
Dominant in	Managed services	Open platforms

Sources

Three capstone agents

Factor	Terminal agent (01)	Issue-to-PR (16)	Migration (09)
Trigger	`agent run <repo> "<task>"`	Issue label `@agent fix this`	`migrate <repo> --target java17`
Sandbox	E2B or Daytona, git worktree	Daytona or E2B per task	Daytona per branch
Turn cap	50 turns	30 agent turns	20 agent turns
Cost cap	$5 per task	$20 per PR, 5 PRs per repo per day	$8 per repo
Other cap	200k context	30 min wall-clock	30 min per repo
Gate before PR	Tests pass	Full CI in sandbox, coverage delta	Tests pass, coverage not down 2%
On failure	`Stop` hook writes a trace bundle	CI red: halt, no PR	Failure class plus repro

Sources

Use X when

Use JSON tool calls when: each action needs an explicit validator and an easy audit.
Use CodeAct when: one action must chain tools over many files. Harden the sandbox first.
Use a terminal agent when: the work is CLI in, pull request out, with a live plan.
Use issue-to-PR when: the UX is a GitHub label. No interactive loop, no terminal.
Use a migration agent when: the test suite is the ground truth across many repos.
Use deterministic recipes first: they catch 70 to 80% of mechanical migration fixes.

Sources

Numbers to know

SWE-bench Verified: 4% to 80.9% in under three years. 500 human-curated tasks.
Scaffold delta: Claude Sonnet 4.5 got 43.2% on SWE-agent v1 and 59.8% on Cline.
Easy tail: 161 of 500 Verified tasks need only a 1 or 2 line change.
SWE-bench Pro: 10+ line changes. Frontier systems score 23 to 59%.
Verifier loop: a 10+ point delta on SWE-bench.
Live-SWE-agent: 79.2% on SWE-bench Verified with Opus 4.5.

Sources

Terminal agent harness (01)

Nodes and connections

Node	Role	Kind
User CLI	agent run	client
Harness	Bun + Ink TUI	process
Plan, act, observe	TodoWrite plan state	process
Model	via OpenRouter	model
Tool dispatcher	MCP StreamableHTTP	process
Sandbox	E2B or Daytona, worktree	process
Hooks	8 lifecycle events	queue
OpenTelemetry	Langfuse: spans, tokens, $	store
PR	via GitHub app	client

From	To	Connection
User CLI	Harness
Harness	Plan, act, observe
Plan, act, observe	Model	calls
Plan, act, observe	Tool dispatcher	tool calls
Tool dispatcher	Sandbox	run
Sandbox	Hooks
Hooks	OpenTelemetry
OpenTelemetry	PR

Sources

phases/19-capstone-projects/01-terminal-native-coding-agent/docs/en.md

Run the lessons

Task	Code
Compare the toy scaffolds (L09)	`code/main.py`
Plan, act, observe loop (01)	`python main.py`
TS harness demo plus offline eval	`npm start`
Simulate 50 migrations (09)	`python main.py`
Dispatch 20 issues over 3 repos (16)	`python main.py`
Fix the fizz fixture (29)	`python3 code/main.py`
Test the end-to-end demo	`python3 -m pytest code/tests/ -v`

Sources

phases/15-autonomous-systems/09-coding-agent-landscape/docs/en.md
phases/19-capstone-projects/01-terminal-native-coding-agent/code
phases/19-capstone-projects/09-code-migration-agent/code/main.py
phases/19-capstone-projects/16-github-issue-to-pr-agent/code/main.py
phases/19-capstone-projects/29-end-to-end-coding-task-demo/docs/en.md

The three projects

Factor	tiny-coding-agent	rust-agent-shell	pr-review-reporter
Builds	Bounded repair loop with exact patches	Rust stdin/stdout action loop	Anchored patch review report
Language	Python 3.10+	Rust plus a Python adapter	TypeScript on Node 22.18+, Python
Input	Workspace plus `proposals.json`	Workspace plus `requests.jsonl`	Unified diff, stdin or `--repo`
Output	`repair-trace.json`	JSONL events per request id	`review.json`, `review.html`, SARIF
Budget	`--max-steps` default 6	`--limit` default 50	Diff at most one megabyte
Safety	Path containment, one test command	No OS shell, 16 KiB reads	HTML escaping, exact quote gate
End states	`completed`, `failed`, `budget_exhausted`	`ok`, `error`, `rejected`	Accepted or rejected findings

Sources

Project commands

Command	What it does
`project_test.py <project> --init learning-artifacts/<project>`	Creates a learner workspace
`project_test.py <project> --stage 1 --path <workspace>`	Grades one stage
`project_test.py <project> --all --solution --strict`	Checks the reference
`cli.py <workspace> <proposals> --copy-to <dir> --out <trace>`	Repairs a disposable copy
`client.py <workspace> <requests> --limit 8 --out <file>`	Rust shell through the adapter
`node cli.ts --diff samples/change.diff --sarif review.sarif`	Writes JSON, HTML and SARIF
`--repo /path --base REF --head REF`	Read-only Git diff mode
`--candidates recorded-findings.json`	Validates another reviewer's findings

Sources

Budgets in the code

Setting	Default	Note
`max_turns`	50	Capstone 01 `Budget`
`max_tokens`	200_000	Capstone 01 context ceiling
`max_dollars`	5.00	Capstone 01, per task
`TRUNCATE_BYTES`	4096	Tool output cap in 01
`BUDGET_TURNS`	20	Migration, with `BUDGET_USD` 8.0
`step_budget`	12	L29 `AgentRun`

Sources

Gate chain and sandbox (29)

chain = GateChain(gates=(

WhitelistGate(allowed=frozenset({"read_file", "write_file", "run_tests"})),allow-set

RegexGate.from_strings(items=(r"\brm\s+-rf\b", r"\bsudo\b", r"^/etc/")),refused patterns

BudgetGate(max_tokens=budget),observation budget

))

outcome = self.chain.evaluate(call, ctx)stops at the first deny

sandbox = Sandbox(project_root=repo, timeout_seconds=10.0)denylist plus path jail

Sources

phases/19-capstone-projects/29-end-to-end-coding-task-demo/code/main.py

Troubleshooting

Symptom	Cause	Fix
Loop crashes on 8MB of ripgrep matches	Untruncated tool output	Cap each call at 4k tokens
`rm -rf` reaches the sandbox	No `PreToolUse` guard	`destructive_guard` sets `blocked`
High Verified score, weak real results	Easy tail, 161 of 500 tasks	Score a Pro-like subset of your backlog
`budget_exhausted`	30 min, $8 or 20 turns used	File the repo with the current diff
`coverage_regression`	Coverage dropped more than 2%	File under the class, no PR
`policy_denied`	`can("force_push")` returns False	Branch protection, app not in bypass list
`patch must match exactly once`	Old text missing or not unique	Use one unique exact old string
`path escapes root`	Path leaves the canonical root	Pass a relative path inside the root

Sources

Q&A

Which coding agent is best?: Wrong question. Measure reliability on your own task distribution.
Why does the toy CodeAct take fewer turns?: One snippet rewrites every failing file at once.
Why no assert for a safety gate?: python -O strips asserts.

Sources

Original source

---
id: coding-agents
title: Coding Agents: Landscape and Capstones
code: CL
subtitle: Pick a scaffold, set the budgets and gates, run the capstone agents and the three coding-agent projects.
hue: 25
sources: phases/15-autonomous-systems/09-coding-agent-landscape, phases/19-capstone-projects/01-terminal-native-coding-agent, phases/19-capstone-projects/09-code-migration-agent, phases/19-capstone-projects/16-github-issue-to-pr-agent, phases/19-capstone-projects/29-end-to-end-coding-task-demo, projects/tiny-coding-agent, projects/rust-agent-shell, projects/pr-review-reporter
source_hash: 409e63299e9dca9c
---

## JSON tool calls vs CodeAct
type: comparison
from: phases/15-autonomous-systems/09-coding-agent-landscape/docs/en.md, phases/15-autonomous-systems/09-coding-agent-landscape/code/main.py

| Factor | JSON tool calls | CodeAct |
|---|---|---|
| One action is | One structured JSON payload | A whole Python program |
| Audit | Easy, one action per turn | Harder, wide blast radius |
| Composition | Limited | Loops, chains tools, catches errors |
| Safety | Validator on every call | Needs a hardened sandbox |
| Toy blast radius | 1 file per action | Every failing file at once |
| Dominant in | Managed services | Open platforms |

## Three capstone agents
type: comparison
span: 2
from: phases/19-capstone-projects/01-terminal-native-coding-agent/docs/en.md, phases/19-capstone-projects/16-github-issue-to-pr-agent/docs/en.md, phases/19-capstone-projects/09-code-migration-agent/docs/en.md

| Factor | Terminal agent (01) | Issue-to-PR (16) | Migration (09) |
|---|---|---|---|
| Trigger | `agent run <repo> "<task>"` | Issue label `@agent fix this` | `migrate <repo> --target java17` |
| Sandbox | E2B or Daytona, git worktree | Daytona or E2B per task | Daytona per branch |
| Turn cap | 50 turns | 30 agent turns | 20 agent turns |
| Cost cap | $5 per task | $20 per PR, 5 PRs per repo per day | $8 per repo |
| Other cap | 200k context | 30 min wall-clock | 30 min per repo |
| Gate before PR | Tests pass | Full CI in sandbox, coverage delta | Tests pass, coverage not down 2% |
| On failure | `Stop` hook writes a trace bundle | CI red: halt, no PR | Failure class plus repro |

## Use X when
type: concepts
from: phases/15-autonomous-systems/09-coding-agent-landscape/docs/en.md, phases/19-capstone-projects/01-terminal-native-coding-agent/docs/en.md, phases/19-capstone-projects/16-github-issue-to-pr-agent/docs/en.md, phases/19-capstone-projects/09-code-migration-agent/docs/en.md

- **Use JSON tool calls when**: each action needs an explicit validator and an easy audit.
- **Use CodeAct when**: one action must chain tools over many files. Harden the sandbox first.
- **Use a terminal agent when**: the work is CLI in, pull request out, with a live plan.
- **Use issue-to-PR when**: the UX is a GitHub label. No interactive loop, no terminal.
- **Use a migration agent when**: the test suite is the ground truth across many repos.
- **Use deterministic recipes first**: they catch 70 to 80% of mechanical migration fixes.

## Numbers to know
type: concepts
from: phases/15-autonomous-systems/09-coding-agent-landscape/docs/en.md, phases/19-capstone-projects/01-terminal-native-coding-agent/docs/en.md

- **SWE-bench Verified**: 4% to 80.9% in under three years. 500 human-curated tasks.
- **Scaffold delta**: Claude Sonnet 4.5 got 43.2% on SWE-agent v1 and 59.8% on Cline.
- **Easy tail**: 161 of 500 Verified tasks need only a 1 or 2 line change.
- **SWE-bench Pro**: 10+ line changes. Frontier systems score 23 to 59%.
- **Verifier loop**: a 10+ point delta on SWE-bench.
- **Live-SWE-agent**: 79.2% on SWE-bench Verified with Opus 4.5.

## Terminal agent harness (01)
type: architecture
from: phases/19-capstone-projects/01-terminal-native-coding-agent/docs/en.md

```arch
node | cli | 1,1 | User CLI | agent run | client
node | harness | 2,1 | Harness | Bun + Ink TUI | process
node | loop | 3,1 | Plan, act, observe | TodoWrite plan state | process
node | model | 4,1 | Model | via OpenRouter | model
node | tools | 3,2 | Tool dispatcher | MCP StreamableHTTP | process
node | box | 2,2 | Sandbox | E2B or Daytona, worktree | process
node | hooks | 1,2 | Hooks | 8 lifecycle events | queue
node | otel | 1,3 | OpenTelemetry | Langfuse: spans, tokens, $ | store
node | pr | 2,3 | PR | via GitHub app | client
edge | cli | harness |
edge | harness | loop |
edge | loop | model | calls
edge | loop | tools | tool calls
edge | tools | box | run
edge | box | hooks |
edge | hooks | otel |
edge | otel | pr |
```

## Run the lessons
type: task-table
span: 2
from: phases/15-autonomous-systems/09-coding-agent-landscape/docs/en.md, phases/19-capstone-projects/01-terminal-native-coding-agent/code, phases/19-capstone-projects/09-code-migration-agent/code/main.py, phases/19-capstone-projects/16-github-issue-to-pr-agent/code/main.py, phases/19-capstone-projects/29-end-to-end-coding-task-demo/docs/en.md

| Task | Code |
|---|---|
| Compare the toy scaffolds (L09) | `code/main.py` |
| Plan, act, observe loop (01) | `python main.py` |
| TS harness demo plus offline eval | `npm start` |
| Simulate 50 migrations (09) | `python main.py` |
| Dispatch 20 issues over 3 repos (16) | `python main.py` |
| Fix the fizz fixture (29) | `python3 code/main.py` |
| Test the end-to-end demo | `python3 -m pytest code/tests/ -v` |

## The three projects
type: comparison
span: 2
col: 1
from: projects/tiny-coding-agent, projects/rust-agent-shell, projects/pr-review-reporter

| Factor | tiny-coding-agent | rust-agent-shell | pr-review-reporter |
|---|---|---|---|
| Builds | Bounded repair loop with exact patches | Rust stdin/stdout action loop | Anchored patch review report |
| Language | Python 3.10+ | Rust plus a Python adapter | TypeScript on Node 22.18+, Python |
| Input | Workspace plus `proposals.json` | Workspace plus `requests.jsonl` | Unified diff, stdin or `--repo` |
| Output | `repair-trace.json` | JSONL events per request id | `review.json`, `review.html`, SARIF |
| Budget | `--max-steps` default 6 | `--limit` default 50 | Diff at most one megabyte |
| Safety | Path containment, one test command | No OS shell, 16 KiB reads | HTML escaping, exact quote gate |
| End states | `completed`, `failed`, `budget_exhausted` | `ok`, `error`, `rejected` | Accepted or rejected findings |

## Project commands
type: command-table
span: 2
col: 3
from: projects/tiny-coding-agent/README.md, projects/rust-agent-shell/README.md, projects/pr-review-reporter/README.md

| Command | What it does |
|---|---|
| `project_test.py <project> --init learning-artifacts/<project>` | Creates a learner workspace |
| `project_test.py <project> --stage 1 --path <workspace>` | Grades one stage |
| `project_test.py <project> --all --solution --strict` | Checks the reference |
| `cli.py <workspace> <proposals> --copy-to <dir> --out <trace>` | Repairs a disposable copy |
| `client.py <workspace> <requests> --limit 8 --out <file>` | Rust shell through the adapter |
| `node cli.ts --diff samples/change.diff --sarif review.sarif` | Writes JSON, HTML and SARIF |
| `--repo /path --base REF --head REF` | Read-only Git diff mode |
| `--candidates recorded-findings.json` | Validates another reviewer's findings |

## Budgets in the code
type: defaults
col: 1
from: phases/19-capstone-projects/01-terminal-native-coding-agent/code/main.py, phases/19-capstone-projects/09-code-migration-agent/code/main.py, phases/19-capstone-projects/29-end-to-end-coding-task-demo/code/main.py

| Setting | Default | Note |
|---|---|---|
| `max_turns` | 50 | Capstone 01 `Budget` |
| `max_tokens` | 200_000 | Capstone 01 context ceiling |
| `max_dollars` | 5.00 | Capstone 01, per task |
| `TRUNCATE_BYTES` | 4096 | Tool output cap in 01 |
| `BUDGET_TURNS` | 20 | Migration, with `BUDGET_USD` 8.0 |
| `step_budget` | 12 | L29 `AgentRun` |

## Gate chain and sandbox (29)
type: annotated-code
col: 1
lang: python
from: phases/19-capstone-projects/29-end-to-end-coding-task-demo/code/main.py

```python
chain = GateChain(gates=(
    WhitelistGate(allowed=frozenset({"read_file", "write_file", "run_tests"})),  # allow-set
    RegexGate.from_strings(items=(r"\brm\s+-rf\b", r"\bsudo\b", r"^/etc/")),  # refused patterns
    BudgetGate(max_tokens=budget),  # observation budget
))
outcome = self.chain.evaluate(call, ctx)  # stops at the first deny
sandbox = Sandbox(project_root=repo, timeout_seconds=10.0)  # denylist plus path jail
```

## Troubleshooting
type: troubleshooting
col: 3
from: phases/15-autonomous-systems/09-coding-agent-landscape/docs/en.md, phases/19-capstone-projects/01-terminal-native-coding-agent/docs/en.md, phases/19-capstone-projects/01-terminal-native-coding-agent/code/main.py, phases/19-capstone-projects/09-code-migration-agent/docs/en.md, phases/19-capstone-projects/16-github-issue-to-pr-agent/docs/en.md, phases/19-capstone-projects/16-github-issue-to-pr-agent/code/main.py, projects/tiny-coding-agent/solution/main.py, projects/rust-agent-shell/solution/main.rs

| Symptom | Cause | Fix |
|---|---|---|
| Loop crashes on 8MB of ripgrep matches | Untruncated tool output | Cap each call at 4k tokens |
| `rm -rf` reaches the sandbox | No `PreToolUse` guard | `destructive_guard` sets `blocked` |
| High Verified score, weak real results | Easy tail, 161 of 500 tasks | Score a Pro-like subset of your backlog |
| `budget_exhausted` | 30 min, $8 or 20 turns used | File the repo with the current diff |
| `coverage_regression` | Coverage dropped more than 2% | File under the class, no PR |
| `policy_denied` | `can("force_push")` returns False | Branch protection, app not in bypass list |
| `patch must match exactly once` | Old text missing or not unique | Use one unique exact old string |
| `path escapes root` | Path leaves the canonical root | Pass a relative path inside the root |

## Q&A
type: qa
span: 2
col: 3
from: phases/15-autonomous-systems/09-coding-agent-landscape/docs/en.md, phases/15-autonomous-systems/09-coding-agent-landscape/code/main.py, phases/19-capstone-projects/16-github-issue-to-pr-agent/code/main.py

- **Which coding agent is best?** Wrong question. Measure reliability on your own task distribution.
- **Why does the toy CodeAct take fewer turns?** One snippet rewrites every failing file at once.
- **Why no `assert` for a safety gate?** `python -O` strips asserts.