docs(entire-cli): enrich 4 guide files with production diagrams from issue #802

6 additions across ai-traceability, ai-ecosystem, third-party-tools, security-hardening.

- ai-traceability.md: full hook architecture diagram (sans/avec Entire),
  real checkpoint structure (entire/checkpoints/v1/ tree), orphan branch
  diagram, Go/No-Go thresholds table with 2h spike commands
- ai-ecosystem.md: agent handoffs flow diagram (Claude → Gemini, no cold start)
- third-party-tools.md: delta table vs existing setups (JSONL, attribution,
  handoffs) + evaluation stop criteria
- security-hardening.md: approval gate flow diagram (policy check →
  low/high risk → approve/reject → audit trail)

Source: github.com/methode-aristote/app/issues/802

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Florian BRUNIAUX 2026-03-05 15:11:32 +01:00
parent 0bdb34b2a4
commit 2b9c654f0f
5 changed files with 180 additions and 8 deletions

View file

@ -1643,6 +1643,34 @@ See: `guide/observability.md` for native Claude Code session monitoring
- ⚠️ Very new (launched Feb 10-12, 2026) - limited production feedback
- 🔗 [GitHub repo](https://github.com/entireio/cli) / [entire.io](https://entire.io)
**Agent handoff flow (how context actually passes between agents):**
```
Claude Code Gemini CLI Entire
----------- ---------- ------
works on feature
"blocked on X,
delegate to Gemini" ---------> hook PreToolUse[Task]
(captures the handoff)
|
receives full |
context: |
- reasoning trace |
- files touched |
- decisions made |
- rejected approaches|
|
works... |
completes |
v
hook PostToolUse[Task]
(captures result)
Result: each agent in the chain sees the reasoning of previous agents.
No "cold start" — full context preserved across agent switches.
```
**When to use Entire CLI:**
```bash

View file

@ -327,6 +327,63 @@ entire init
entire capture --agent="claude-code"
```
**How It Works (Hook Architecture):**
```
WITHOUT ENTIRE
==============
Developer Agent (Claude/Gemini/Codex) Git
--------- --------------------------- ---
prompt ---------> reasons + edits files
tool calls (Bash, Read, Edit...)
prompt ---------> continues...
"looks good" ---> session ends
git commit -----> ----------------------------------------> commit on feature/branch
(code only, zero context)
Result: the code is there, but WHY and HOW are lost.
No record of prompts, reasoning, or abandoned approaches.
WITH ENTIRE
===========
Developer Agent (Claude/Gemini/Codex) Entire Hooks Git
--------- --------------------------- ------------ ---
entire enable ---> installs 7 hooks automatically (once per repo)
[SESSION START] -----------------------------------------> hook SessionStart
prompt ---------> reasons + edits ---------> hook UserPromptSubmit
tool calls... ---------> hook PreToolUse/PostToolUse
[AGENT ENDS] -------------------------------------------------> hook Stop
|
CHECKPOINT created on
shadow branch:
entire/2b4c177-a5e3f2
|
Contains:
- full transcript
- user prompts
- file diffs
- tool calls
- token usage
- human vs AI attribution %
git commit -----> ----------------------------------------> commit on feature/branch
+ auto-added trailer:
"Entire-Checkpoint: a3b2c4"
git push -------> ----------------------------------------> code pushed normally
shadow → entire/checkpoints/v1
(orphan branch, zero conflicts)
shadow branch auto-deleted
```
**Workflow with Claude Code:**
```bash
@ -409,18 +466,26 @@ entire approve --approver="jane@company.com"
**Architecture:**
Entire stores data in `.entire/` directory with separate git branch:
Entire stores checkpoints on an orphan branch — no common ancestor with `main`, so no merge conflicts and no history pollution:
```
project/
├─ .entire/
│ ├─ config.yaml # Configuration
│ ├─ sessions/ # Session metadata
│ └─ checkpoints/ # Named checkpoints
└─ .git/
└─ refs/heads/entire/checkpoints/v1
entire/checkpoints/v1/ ← orphan branch (no common ancestor with main)
├─ a/b2c4d5e6f7/ ← checkpoint ID (random hex)
│ ├─ metadata.json ← summary, attribution %, token count
│ └─ 0/
│ ├─ full.jsonl ← complete session transcript
│ ├─ prompt.txt ← user prompts
│ └─ context.md ← generated context summary
└─ c/d4e5f6a7b8/ ← another checkpoint
└─ ...
main ----o----o----o----o----> (normal code history, untouched)
entire/checkpoints/v1 ----x----x----x----> (no common ancestor = no merge conflicts)
```
Why orphan branch: `git clone --single-branch` ignores checkpoints (zero overhead for consumers). Multiple devs can push in parallel without conflicts (checkpoint IDs are unique).
**Limitations:**
- Very new (launched Feb 10-12, 2026) - limited production feedback
@ -436,6 +501,33 @@ project/
- ✅ Governance gates (approval required before actions)
- ⚠️ Personal projects: May be overkill (simple `Co-Authored-By` suffices)
**Go/No-Go evaluation thresholds (run a 2h spike before team rollout):**
```bash
# Install on a throwaway branch
entire enable
# After 2-3 normal sessions, measure:
du -sh .git/refs/heads/entire/ # Storage overhead per session
time git push # Push time including condensation
ls .git/hooks/ # Check for conflicts with existing hooks
```
| Metric | Green (proceed) | Red (stop) |
|--------|----------------|-----------|
| Checkpoint size | < 10 MB/session | > 10 MB → storage risk |
| Push overhead | < 5s | > 5s → daily friction |
| Repo growth | < 100 MB/week | > 100 MB/week |
| Hook compatibility | No conflicts | Timeout or conflict → blocker |
**Team size guidance:**
| Team | Recommendation |
|------|---------------|
| Solo dev | `Co-Authored-By` trailer suffices |
| 2-5 devs | Justified if multi-agent workflows or shared audit trail needed |
| 5+ devs / enterprise | Strong fit (shared checkpoints, governance, compliance) |
### 5.2 Automated Attribution Hook
Add `Assisted-by` trailer automatically when Claude Code commits:

View file

@ -578,6 +578,28 @@ The agent checks:
| **Approvals** | Approver identity, timestamp, checkpoint reference | Permanent |
| **Agent handoffs** | Source/target agents, context transferred | Permanent |
**Approval gate flow:**
```
Developer --> commit + checkpoint
|
v
[Policy Check]
"Does this touch prisma/schema.prisma?"
"Does this touch src/server/auth*?"
|
+----+----+
| |
Low risk High risk
| |
Auto-OK Approval Gate
"Reviewer inspects:
transcript + diffs + attribution %"
|
Approve / Reject
(immutable audit trail entry)
```
**Example compliance workflow:**
```bash

View file

@ -250,6 +250,30 @@ Agent-native platform for Git-integrated session capture with rewindable checkpo
- macOS/Linux only (Windows via WSL)
- Early stage (v1.x) - expect API changes
**Delta vs common existing setups:**
| Need | Typical existing setup | What Entire adds |
|------|----------------------|-----------------|
| Tool call logging | Local JSONL (7-day rotation) | Reasoning + attribution %, Git-permanent |
| Human/AI attribution | Nothing | % per file, annotated per line, by model |
| Agent handoffs | Manual context copy | Context checkpoint auto-passed to next agent |
| Inter-dev handoff | Git commits/PRs | Shared readable checkpoints on `entire/checkpoints/v1` |
| Session persistence | Local only, ephemeral | Git-native, permanent, shareable |
| Governance | Custom pre-commit hooks | Policy-based approval gates + configurable audit export |
**Evaluation (2h spike recommended before team rollout):**
```bash
entire enable # Install on throwaway branch
# After 2-3 normal sessions:
du -sh .git/refs/heads/entire/ # Storage per session → flag if > 10 MB
time git push # Push overhead → flag if > 5s
ls .git/hooks/ # Verify no conflict with existing hooks
```
Stop criteria: checkpoint > 10 MB/session, push overhead > 5s, or hook conflicts.
> **Cross-ref**: Full Entire workflow with examples at [AI Traceability Guide](./ai-traceability.md#51-entire-cli). For compliance use cases, see [Security Hardening](./security-hardening.md).
---