# Context Analysis — 2026-07-11 > Source: `/wr-retrospective:analyze-context` (deep layer per ADR-043). > Methodology: byte-count-on-disk + per-plugin decomposition (cache-fallback) + per-turn attribution (when session log available). > Cheap-layer baseline: `packages/retrospective/scripts/measure-context-budget.sh`. > Auto-fired from run-retro Step 2c: first measurement for this project (no prior context-analysis report existed → calendar-elapse trigger held). ## Bucket Totals Project on-disk surfaces (cheap-layer buckets). `hooks` / `skills` are source-absent in this adopter tree (they live in the plugin cache, not the project) — decomposed separately below via cache-fallback. | Bucket | Bytes | % of measured | Δ vs prior | | ------------------ | -------------------------------------------------------------------------------- | ------------- | -------------------------------------------------- | | memory | 572,996 | 46.2% | no prior snapshot — first measurement this project | | decisions | 299,777 | 24.2% | no prior snapshot — first measurement this project | | problems | 274,088 | 22.1% | no prior snapshot — first measurement this project | | jtbd | 46,575 | 3.8% | no prior snapshot — first measurement this project | | briefing | 38,968 | 3.1% | no prior snapshot — first measurement this project | | project-claude-md | 7,747 | 0.6% | no prior snapshot — first measurement this project | | hooks | not measured — source-absent (plugin-cache, not project tree; see decomposition) | — | — | | skills | not measured — source-absent (plugin-cache, not project tree; see decomposition) | — | — | | framework-injected | not measured — framework-injected-no-on-disk-source | — | — | Total measured (project on-disk buckets): **1,240,151 bytes**. Threshold (cheap-layer per-bucket ceiling): 10,240 bytes — every measured project bucket except `project-claude-md` is over the advisory ceiling, expected for a mature governance corpus (`decisions` / `problems` / `jtbd` grow monotonically by design). ## Per-Plugin Decomposition Resolved via `wr-retrospective-list-plugin-attribution` cache-fallback mode (no `packages/` in this adopter tree; helper sniffed plugin-cache `bin/` entries). ### Hooks (aggregate: 519,395 bytes) | Plugin | Bytes | % of hooks | | -------------------- | ------- | ---------- | | wr-itil | 186,931 | 36.0% | | wr-risk-scorer | 122,270 | 23.5% | | wr-architect | 68,821 | 13.2% | | wr-voice-tone | 59,564 | 11.5% | | wr-jtbd | 39,893 | 7.7% | | wr-style-guide | 24,305 | 4.7% | | wr-retrospective | 17,163 | 3.3% | | ponytail | 448 | 0.1% | | accessibility-agents | 0 | 0.0% | ### Skills (aggregate: 1,311,265 bytes) | Plugin | Bytes | % of skills | | ---------------- | --------- | ----------- | | wr-itil | 1,004,794 | 76.6% | | wr-retrospective | 116,481 | 8.9% | | wr-risk-scorer | 74,674 | 5.7% | | wr-architect | 64,100 | 4.9% | | wr-jtbd | 21,702 | 1.7% | | ponytail | 15,500 | 1.2% | | wr-voice-tone | 10,119 | 0.8% | | wr-style-guide | 3,895 | 0.3% | ## Top-N Offenders | Surface | Bytes | Bucket | Comparable prior | | -------------------------------- | --------- | --------- | ----------------------------- | | wr-itil skills (plugin cache) | 1,004,794 | skills | not estimated — no prior data | | memory (`~/.claude/.../memory/`) | 572,996 | memory | not estimated — no prior data | | decisions (`docs/decisions/`) | 299,777 | decisions | not estimated — no prior data | | problems (`docs/problems/`) | 274,088 | problems | not estimated — no prior data | | wr-itil hooks (plugin cache) | 186,931 | hooks | not estimated — no prior data | ## Per-Turn Attribution per-turn attribution: not measured — no session log accessible (only `.afk-run-state/outstanding-questions.jsonl` and `risk-register-queue.jsonl` present; neither carries per-turn `usage` fields). ## Suggestions Per ADR-026 — this is the first measurement, so no comparable prior byte-saving data exists for this project; approaches are cited from the wr-retrospective plugin's own reclamation history, byte estimates marked ungrounded. 1. **skills / wr-itil (1,004,794 bytes)** — The `wr-itil` SKILL corpus dominates loaded context; the largest single file is `manage-problem/SKILL.md` (~52 KB observed this session, exceeding the ADR-038 P097 50 KB cluster anchor). Approach precedent: `P100` split the monolithic `BRIEFING.md` into per-topic files loaded on demand; the same lazy-load / REFERENCE.md-split pattern (per ADR-054) applies to oversized SKILL bodies. This is an upstream `@windyroad/itil` concern, not a dry-aged-deps surface. Estimated byte saving: not estimated — no prior data. 2. **memory (572,996 bytes)** — The per-project auto-memory corpus is the second-largest bucket. Approach precedent: `P105` signal-vs-noise scoring drives decay-based removal of unused entries. Estimated byte saving: not estimated — no prior data. 3. **problems (274,088 bytes)** — `docs/problems/README.md` line-3 accumulator was already bounded by `P134` (README-history rotation). Continued closed-ticket archival keeps the active queue readable. Estimated byte saving: not estimated — no prior data. ## Policy Breaches no policy breaches detected on project surfaces measured. ADR-040 Tier-3 briefing budget check (`check-briefing-budgets.sh`) not resolvable in this adopter tree (no `packages/`); the run-retro Step 3 Tier-3 pass is the load-bearing surface for briefing budgets.