feat(#270): /threat-model inlines DFD as point-in-time snapshot#273
Merged
Conversation
…dit time The threat-model audit was linking to the live DFD instead of snapshotting it. So when /dfd was re-run after architecture changes, the historical threat models silently referenced a DFD that no longer matched what they analysed — last quarter's threats indexed against this quarter's DFD. Changes: - Step 1 now REFUSES if `projects/<name>/architecture/dfd.md` is missing (was: degraded "inline discovery" fallback). A threat model without a DFD has no point-in-time anchor; future readers can't replay the analysis. Fail fast, prompt the operator to run /dfd first. - New Step 1b extracts three sections from dfd.md into shell variables: the ```mermaid ``` block, the `## Trust boundaries` table, and the `## Data classifications` table. Discovery provenance is intentionally excluded (too noisy for an audit snapshot). - Step 5b's body template now embeds a `## DFD (snapshot as of YYYY-MM-DD)` section at the top, with a callout note pointing at the live file (`projects/<name>/architecture/dfd.md`) for "current state". - After audit_run_persist, the audit MD is passed through the shared Mermaid lint (`_lib-mermaid-lint.sh` from #266) — if the live DFD has broken Mermaid, the snapshot inherits it; better to surface here than on GitHub. - Rules + Anti-patterns sections updated with the snapshot rationale. - /dfd SKILL.md gets a one-line forward-link explaining the audit consumers snapshot the file at audit time. Refs #270 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
atlas-apex
commented
May 17, 2026
atlas-apex
left a comment
Collaborator
Author
There was a problem hiding this comment.
Code Review: PR #273
Commit: 9950c1a2bc84100f442b858041b8dc2d1ca1d63f
Summary
SKILL.md-only change. /threat-model now refuses on missing DFD (no fallback), extracts three sections from dfd.md (Mermaid block, Trust boundaries, Data classifications) in a new Step 1b, and inlines them at the top of the audit MD as a point-in-time snapshot. Step 5b heredoc switched from quoted to unquoted EOF to interpolate the captured variables. Post-persist Mermaid lint added. One-line forward-link in /dfd SKILL.md explains that editing the live DFD does not invalidate prior audit snapshots.
Checklist Results
- Architecture & Design: Pass — preserves AgDR-0019 dated-subdir persistence; AgDR-0026 (DFD as source of truth) honoured by the refusal-not-fallback choice
- Code Quality: Pass — awk extraction patterns are standard; heredoc quoting switch correctly enables
${dfd_*}interpolation - Testing: N/A — markdown-driven skill, no shell logic that ships; PR body lists fence-balance + Step-1-block + heredoc-interpolation manual checks
- Security: Pass — no secrets, no external calls, no auth-adjacent paths
- Performance: Pass — three awk passes over a small file at audit time; trivial
- PR Description & Glossary: Pass — Glossary has 4 terms (Snapshot, Live DFD, Audit-artefact persistence, Refusal vs fallback)
- Technical Decisions (AgDR):N/A — behaviour change to existing skill; AgDR-0019 referenced for persistence; design rationale captured inline in SKILL.md "Implementation notes" #8–#10 + new "Anti-patterns" section. No new architectural decision.
- Adopter Handbooks: N/A — diff is markdown-only, no language handbooks triggered
Issues Found
None.
Suggestions
- nit: the two awk extractors for
## Trust boundaries/## Data classificationsuse&& !/^## Trust//&& !/^## Data classifications/to avoid exiting on the capture line. Works given section ordering in the/dfdtemplate; a slightly more robust pattern would becapture && /^## / && !/^## (Trust boundaries|Data classifications)/ { exit }after the capture flag flips. Optional. - nit:
audit_md="$(audit_resolve_dir ...)/${ts//:/}.md"assumesaudit_resolve_diris in scope at Step 5b — worth confirming_lib-audit-persist.shexports it (or sourcing the lib in the snippet). End-to-end operator verification in the PR body covers this.
Verdict
APPROVED (comment review — author cannot self-approve via GitHub UI)
The point-in-time snapshot pattern is the right call — it makes the dated audits in audits/threat-model/ independently re-readable later, which is the whole point of the dated-subdir convention from AgDR-0019. The refusal-not-fallback design is well-justified in SKILL.md and the PR body. Glossary present. Fence balance verified (28 fences = 14 pairs).
🤖 Reviewed by Rex (Code Reviewer Agent)
📌 Reviewed commit: 9950c1a2bc84100f442b858041b8dc2d1ca1d63f
This was referenced May 17, 2026
Merged
me2resh
added a commit
that referenced
this pull request
Jun 5, 2026
…dit time (#273) The threat-model audit was linking to the live DFD instead of snapshotting it. So when /dfd was re-run after architecture changes, the historical threat models silently referenced a DFD that no longer matched what they analysed — last quarter's threats indexed against this quarter's DFD. Changes: - Step 1 now REFUSES if `projects/<name>/architecture/dfd.md` is missing (was: degraded "inline discovery" fallback). A threat model without a DFD has no point-in-time anchor; future readers can't replay the analysis. Fail fast, prompt the operator to run /dfd first. - New Step 1b extracts three sections from dfd.md into shell variables: the ```mermaid ``` block, the `## Trust boundaries` table, and the `## Data classifications` table. Discovery provenance is intentionally excluded (too noisy for an audit snapshot). - Step 5b's body template now embeds a `## DFD (snapshot as of YYYY-MM-DD)` section at the top, with a callout note pointing at the live file (`projects/<name>/architecture/dfd.md`) for "current state". - After audit_run_persist, the audit MD is passed through the shared Mermaid lint (`_lib-mermaid-lint.sh` from #266) — if the live DFD has broken Mermaid, the snapshot inherits it; better to surface here than on GitHub. - Rules + Anti-patterns sections updated with the snapshot rationale. - /dfd SKILL.md gets a one-line forward-link explaining the audit consumers snapshot the file at audit time. Refs #270 Co-authored-by: me2resh <ahmed.abdelaliem@gmail.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Closes the silent-rot problem in historical threat models: after
/dfdis re-run on a project, the previously-written threat models were referencing a DFD that no longer matched what they had analysed. Now/threat-modelsnapshots the DFD into its audit artefact at audit time — the live DFD can evolve freely after, and the snapshot stays frozen.What changed
dfd.mdwas missingdfd.mdby pathdfd.md: the Mermaidflowchartblock, the## Trust boundariestable, the## Data classificationstable_lib-mermaid-lint.sh(from #266) — if the live DFD has broken Mermaid, the snapshot inherits it; better to surface here than on GitHubWhere the snapshot lands
Per AgDR-0019, the audit artefact still writes to
projects/<name>/audits/threat-model/<ts>.md+runs/<ts>.json. The change is purely to the body content of the MD — the dated-subdir history convention is unchanged.How the audit reads later
A threat-model from May references the DFD as it was in May. A threat-model from August references the DFD as it was in August. The diff between two dated audits in
audits/threat-model/now includes BOTH the threats AND the architecture the threats were enumerated against — making "what changed" legible across runs.Testing
```fence count is balanced (14 open / 14 closed)exit 1after the BLOCKED message; no fallback path remains<<'EOF'(no expansion) to<<EOF(expansion) so${dfd_mermaid}/${dfd_trust}/${dfd_classifications}/${dfd_captured_at}interpolate correctly/threat-model <registered project>after this merges; confirm the snapshot section renders)Out of scope (deferred to follow-up tickets if useful)
/compliance-check— also readsdfd.md; same rot problem; file as #270b when ready/launch-check— surface staleness at portfolio-level.Closes #270.
Glossary
projects/<name>/architecture/dfd.mdproduced by/dfdprojects/<name>/audits/<dim>/<ts>.md